diff --git a/images/neon/ndm-01-cam-dynamic_obj-static-death/follow.png b/images/neon/ndm-01-cam-dynamic_obj-static-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..c4c495de8aed1cc6a92f98867082c7e9b4e72207
--- /dev/null
+++ b/images/neon/ndm-01-cam-dynamic_obj-static-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2709e3b69bffdf9c269f242f1dd27a291775041e500bbe8e10d86e66748798d6
+size 442575
diff --git a/images/neon/ndm-02-cam-dynamic_obj-static-death/follow.png b/images/neon/ndm-02-cam-dynamic_obj-static-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..23f9c2b5defe6f2380839188b6977905da540dba
--- /dev/null
+++ b/images/neon/ndm-02-cam-dynamic_obj-static-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:faa913209274f359c19fdfd3f765a732aba0200a9d98b31f75013925967154db
+size 466330
diff --git a/images/neon/ndm-02-cam-static_obj-dynamic-death/follow.png b/images/neon/ndm-02-cam-static_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..26ff089bd21d1089baec13bd2be31bfb580306e9
--- /dev/null
+++ b/images/neon/ndm-02-cam-static_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d0159ce3a3ed6f127600e55003da3132000d8b6b7abcf0da901ff2acac6374c
+size 471017
diff --git a/images/neon/ndm-02-cam-static_obj-enter-death/follow.png b/images/neon/ndm-02-cam-static_obj-enter-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..7bdfcae48847dbfa5528007bda3e82fea08742f0
--- /dev/null
+++ b/images/neon/ndm-02-cam-static_obj-enter-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7778d7f24f5ccf7ef861db0ddaf7923fdf9ce004a994d03e73cf6e59b5bea64
+size 464859
diff --git a/images/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow.png b/images/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..541afd8ce0f4178b8c00ac7e2da4cc3d8418b391
--- /dev/null
+++ b/images/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c29e9889c5feb5ad5b39881bdb8d3fcb02dc620964fd986a33f835e135f9070
+size 420635
diff --git a/images/neon/ndm-03-cam-static_obj-dynamic-death/follow.png b/images/neon/ndm-03-cam-static_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..003b29c400afd0cc397d8666c5e9663e07093c40
--- /dev/null
+++ b/images/neon/ndm-03-cam-static_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d542b2832c56904973c5491d436ca24f1dcf4d55fc61cc10700c500f9bf5fd56
+size 450912
diff --git a/images/neon/ndm-04-cam-dynamic_obj-static-death/follow.png b/images/neon/ndm-04-cam-dynamic_obj-static-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..f59511e03a86edce9366117b9da17f1585d391d4
--- /dev/null
+++ b/images/neon/ndm-04-cam-dynamic_obj-static-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e5ce244fd019e710b6d6d8400bebe85f2c7612c8a9136efec4277984a39ac25
+size 413737
diff --git a/images/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow.png b/images/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..be2c6b86eb78dfd6d9e1792091af73552c8645de
--- /dev/null
+++ b/images/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fc9eb3336f5786b152c79f07fcb10622ecb1357728d77bc348a486d5caa5892
+size 461044
diff --git a/images/neon/ndm-05-cam-static_obj-dynamic-death/follow.png b/images/neon/ndm-05-cam-static_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..a23cd731b3685479f867e6636b902bc84da60f35
--- /dev/null
+++ b/images/neon/ndm-05-cam-static_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcd80a1337b96660e9a884e1c186729e409350ac9238742275fd4427123ebb46
+size 473717
diff --git a/images/neon/ndm-05-cam-static_obj-enter-death/follow.png b/images/neon/ndm-05-cam-static_obj-enter-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..fe47c3d16a84f641779ef7698f2947b0df6b1570
--- /dev/null
+++ b/images/neon/ndm-05-cam-static_obj-enter-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:473fa5dbd0921affe9444c3a77643c1cadd6ab274008f88cc790bb200b9dfe0c
+size 484320
diff --git a/images/neon/ndm-06-cam-static_obj-dynamic-death/follow.png b/images/neon/ndm-06-cam-static_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..c70abf28a9ee1bfbaa5467cd0467f5a1d6444263
--- /dev/null
+++ b/images/neon/ndm-06-cam-static_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a7d7e18b2bc5af4761a83574ab8adf86f07499d5f7fa29b01fbab7a365dddf5
+size 438824
diff --git a/images/neon/ndm-06-cam-static_obj-static-death/follow.png b/images/neon/ndm-06-cam-static_obj-static-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..945a3a1a451c863a55e0952c1dc137661ddb0760
--- /dev/null
+++ b/images/neon/ndm-06-cam-static_obj-static-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca3c25558416bc6a48a8d443efcb6757d13a4137983e93e9602b77261ba74b18
+size 444575
diff --git a/images/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow.png b/images/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..c54c049d62c0af66b0cb64070182013c25f34f47
--- /dev/null
+++ b/images/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fdfe7173d2429d2ecaf530b6c65175ec7bcfd770534456f185021e7ca6a0c98
+size 437128
diff --git a/images/neon/ndm-08-cam-dynamic_obj-dynamic-death/follow.png b/images/neon/ndm-08-cam-dynamic_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..153cf8c2abc5032cbf147fb6a5520db4a38b5988
--- /dev/null
+++ b/images/neon/ndm-08-cam-dynamic_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6cfe2a1cd1c779febe81dee610abc10b36bef1197501f3f40af419fc3a0bdee
+size 491449
diff --git a/images/neon/ndm-08-cam-static_obj-dynamic-death/follow.png b/images/neon/ndm-08-cam-static_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..82681c8f36c01dee3cbf0d1aa32ee72588e175a9
--- /dev/null
+++ b/images/neon/ndm-08-cam-static_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59597cd5651a912eeb8803c0e42ccf12333576bdbd8068e630006d0e1235d1fb
+size 464418
diff --git a/images/neon/ndm-08-cam-static_obj-static-death/follow.png b/images/neon/ndm-08-cam-static_obj-static-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..937e5c65635de4ccb44bad54bb9e95454563397f
--- /dev/null
+++ b/images/neon/ndm-08-cam-static_obj-static-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:922fe4fb5f79c0c4676b8e94f56dbfb4489ef2aecdf3170db126e5a2c075b917
+size 476055
diff --git a/images/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow.png b/images/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..c4464d80a43896c6df3042b40a90edce4a93dfba
--- /dev/null
+++ b/images/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3b28a03082406e953e5b36474c68b5cc3dc85a41eefd2ca0471ccbb791a55e9
+size 437103
diff --git a/images/neon/ndm-09-cam-static_obj-enter-death/follow.png b/images/neon/ndm-09-cam-static_obj-enter-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..24fb5f0210b517d5a3a8ec6e7fea3a9a19c8217d
--- /dev/null
+++ b/images/neon/ndm-09-cam-static_obj-enter-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:635d8660d19bffd8c9c11d463f2952a03c60bc867ace20ace495a1b79381a85e
+size 453052
diff --git a/images/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow.png b/images/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..ea5459ccf62d870386d649cd7f2ca077b7999802
--- /dev/null
+++ b/images/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98be93ce7e438e2882d9585cc9b86f84e653da7fb468e124f0a757fc55db73bb
+size 342736
diff --git a/images/neon/ndm-10-cam-static_obj-dynamic-death/follow.png b/images/neon/ndm-10-cam-static_obj-dynamic-death/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..3aa7505aabefc0c27a5da655e9a4e5566409993c
--- /dev/null
+++ b/images/neon/ndm-10-cam-static_obj-dynamic-death/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a434113c75e6e6694c251ef86470e9693e4ca8842a4e9406142025a9ed118d1
+size 465824
diff --git a/images/verified_first_frame/dv200-001-city-crossing/follow.png b/images/verified_first_frame/dv200-001-city-crossing/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..a0d5adcdf22fac0e6a1eb9b3d86f265eccdab38d
--- /dev/null
+++ b/images/verified_first_frame/dv200-001-city-crossing/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7888697eb2f8368fcd7e7ae2a7c5dcab9263d0dc6e268c6e8ed93927a952f682
+size 877931
diff --git a/images/verified_first_frame/dv200-002-city-drive/follow.png b/images/verified_first_frame/dv200-002-city-drive/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..e3dc5af1c6ea442a97be6c2fbe14a3cb9d46ba86
--- /dev/null
+++ b/images/verified_first_frame/dv200-002-city-drive/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c1722b584273bb9d6cc1050564487e28b897f1e8558be010cdadcae294c0d57
+size 1069978
diff --git a/images/verified_first_frame/dv200-006-city-handoff/follow.png b/images/verified_first_frame/dv200-006-city-handoff/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..43540fd8f7e5f261db7a279493b66ce453ccb7ae
--- /dev/null
+++ b/images/verified_first_frame/dv200-006-city-handoff/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e15babe534251d4a4de43b81e7055d9ef62f9f1df7e1fb61c02bba5162e0130
+size 1097323
diff --git a/images/verified_first_frame/dv200-008-city-cart/first-person.png b/images/verified_first_frame/dv200-008-city-cart/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..c80a13a9b3d5a92f39be3f349aa0821a9d9386f6
--- /dev/null
+++ b/images/verified_first_frame/dv200-008-city-cart/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d95b2c39f35adfa03c8544a808a25fe0827f24f3b364a32c010f74f41c8039b8
+size 1144069
diff --git a/images/verified_first_frame/dv200-014-city-push/follow.png b/images/verified_first_frame/dv200-014-city-push/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..45931d8a36b025f0a9279dfe36ba0b165e7f7806
--- /dev/null
+++ b/images/verified_first_frame/dv200-014-city-push/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2056d3ad73201f34796d8fb9d8737074476788f8d3e191a4df2e6f6e02a90275
+size 900338
diff --git a/images/verified_first_frame/dv200-023-indoor-sit/follow.png b/images/verified_first_frame/dv200-023-indoor-sit/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..0e001de635567597afecca8e274e0b42f4a69809
--- /dev/null
+++ b/images/verified_first_frame/dv200-023-indoor-sit/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ec09c100cf0be05b0e08f2f1ed07edb7ec1c5ce663ca146a2ec23d6ce8a8fa8
+size 1129999
diff --git a/images/verified_first_frame/dv200-028-indoor-crowd/follow.png b/images/verified_first_frame/dv200-028-indoor-crowd/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..13cb022c811886c941d13b3d067090040f3de92c
--- /dev/null
+++ b/images/verified_first_frame/dv200-028-indoor-crowd/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:517c7ba1a0696842730a87b44fde5a43058e9351d7386cc9340e8dc53c363b0d
+size 879940
diff --git a/images/verified_first_frame/dv200-030-indoor-cart/follow.png b/images/verified_first_frame/dv200-030-indoor-cart/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..eecc21caccdd5b14b55656e32900b6451129f461
--- /dev/null
+++ b/images/verified_first_frame/dv200-030-indoor-cart/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df4ecdabfd4b801b5c48f951fc589efc8c7e54acf776f780eb5a5f4d3465d382
+size 1019755
diff --git a/images/verified_first_frame/dv200-033-indoor-carryStairs/follow.png b/images/verified_first_frame/dv200-033-indoor-carryStairs/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..3cf1f30af9cafeb47815b6a6f40bf4d5fc55e6d4
--- /dev/null
+++ b/images/verified_first_frame/dv200-033-indoor-carryStairs/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c10c6546def446af14bf04e7cde17ce6586f8e265522de831eb75e3a89688069
+size 759575
diff --git a/images/verified_first_frame/dv200-051-nature-slope/follow.png b/images/verified_first_frame/dv200-051-nature-slope/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..c2ffa256e3f3062d2518e5d2992bf809529dd86e
--- /dev/null
+++ b/images/verified_first_frame/dv200-051-nature-slope/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fad94783a11accdb84008d3b5c3106f4eaf7c041de6637322fb85f28392f0244
+size 876707
diff --git a/images/verified_first_frame/dv200-056-nature-slide/follow.png b/images/verified_first_frame/dv200-056-nature-slide/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..81d676bbd1aff39009544abaab48dbd5c05d0555
--- /dev/null
+++ b/images/verified_first_frame/dv200-056-nature-slide/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a97a9dd036ffcb7d7fcb3f887e6a500f86193cb3f91b2f6db30cd8d3920788cc
+size 791176
diff --git a/images/verified_first_frame/dv200-074-sky-helicopter/first-person.png b/images/verified_first_frame/dv200-074-sky-helicopter/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..4dfe8a2f618256fa372be4239bf71549c8ab5bc9
--- /dev/null
+++ b/images/verified_first_frame/dv200-074-sky-helicopter/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c97315dc796f758446f309b8cba2bfff670301a77e0ff476b428eb8f2a746f3
+size 801535
diff --git a/images/verified_first_frame/dv200-079-racing-racing/follow.png b/images/verified_first_frame/dv200-079-racing-racing/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..9b5e70b6e2c519623b9f0cd8f55ab26cec0139b3
--- /dev/null
+++ b/images/verified_first_frame/dv200-079-racing-racing/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0335a5b3180ebab808e128c8562b57f0396c88a0af4548a141dbc0b2c425141
+size 1033244
diff --git a/images/verified_first_frame/dv200-084-racing-racing/follow.png b/images/verified_first_frame/dv200-084-racing-racing/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..5266acb1a934c7f2207029339abd3568db837c88
--- /dev/null
+++ b/images/verified_first_frame/dv200-084-racing-racing/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f3f62c452e64f6068cdfc7f2f9ac1964414bdbf2b4f149e91d6d077609ef5cd
+size 637309
diff --git a/images/verified_first_frame/dv200-085-racing-racing/follow.png b/images/verified_first_frame/dv200-085-racing-racing/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..4ddf24c5527179c2ba90c9a7b24ab5e92bcfd694
--- /dev/null
+++ b/images/verified_first_frame/dv200-085-racing-racing/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2fcf4935fe3b19ba04f1989579080148c4c636558c04b08d118ee9386369e980
+size 706232
diff --git a/images/verified_first_frame/dv200-091-fantasy-walk/first-person.png b/images/verified_first_frame/dv200-091-fantasy-walk/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..7eb19ddefde16e2984482788ea97a10223aeec78
--- /dev/null
+++ b/images/verified_first_frame/dv200-091-fantasy-walk/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f5fccbb1a496fd3f92198dce9ec0ace72a0e32d57440a909dd5421d2a713ffb
+size 853508
diff --git a/images/verified_first_frame/dv200-094-fantasy-walk/follow.png b/images/verified_first_frame/dv200-094-fantasy-walk/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..9ec78b259e1a0498ee9ce87ea9641d55f71ef5d9
--- /dev/null
+++ b/images/verified_first_frame/dv200-094-fantasy-walk/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8477ee58929521e19afddba7d9c5b4814f3d4f4d6f500e0a80c2220af6479f40
+size 1118869
diff --git a/images/verified_first_frame/dv200-100-fantasy-walk/follow.png b/images/verified_first_frame/dv200-100-fantasy-walk/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..6522723fc3b4728762f6d35ba8f41329dd4a4576
--- /dev/null
+++ b/images/verified_first_frame/dv200-100-fantasy-walk/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2430ee699335a82ddfbec7b213f956bc8779def2aa7ecd7e92e84e053c3c2105
+size 972273
diff --git a/images/verified_first_frame/dv200-103-city-traffic/follow.png b/images/verified_first_frame/dv200-103-city-traffic/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..52075924c96aeb7e2fa569a5325cccfd760c748a
--- /dev/null
+++ b/images/verified_first_frame/dv200-103-city-traffic/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:187b04414af3da0d529938edbe3e1b5d9dddc1f87c49b14b4d846d8957f4505c
+size 808494
diff --git a/images/verified_first_frame/dv200-107-city-stairsDown/follow.png b/images/verified_first_frame/dv200-107-city-stairsDown/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..cd1d17848f96c5e0c7c359612aaf1bd3ccfa2276
--- /dev/null
+++ b/images/verified_first_frame/dv200-107-city-stairsDown/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a166824ac12d0c6c340088302e3ca4ed4023b99e6e958e7d16bd700875c1a21
+size 951627
diff --git a/images/verified_first_frame/dv200-114-city-push/follow.png b/images/verified_first_frame/dv200-114-city-push/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..148410a42613e6145984b7588c41c09a03a83b3c
--- /dev/null
+++ b/images/verified_first_frame/dv200-114-city-push/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e00ec6c68c6236b994005170b5dc8be282526a9b6e29dac5023e3a637853578c
+size 721410
diff --git a/images/verified_first_frame/dv200-123-indoor-sit/follow.png b/images/verified_first_frame/dv200-123-indoor-sit/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..5cf9a0edf26994fc630913c31fa7e76adc2566f4
--- /dev/null
+++ b/images/verified_first_frame/dv200-123-indoor-sit/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b22e0dc1b87e0820a211c070d3d766b9f69ab6501fb62ea7538d1441b6e4b1d1
+size 634973
diff --git a/images/verified_first_frame/dv200-126-indoor-stairs/follow.png b/images/verified_first_frame/dv200-126-indoor-stairs/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..3402b5de3b45912c67a33f4a07ce7c03c740ff35
--- /dev/null
+++ b/images/verified_first_frame/dv200-126-indoor-stairs/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1cc8418c24774fed5a84b756b48f4574b3d2d81b504b2accfe62d6d91b23386
+size 862785
diff --git a/images/verified_first_frame/dv200-130-indoor-cart/follow.png b/images/verified_first_frame/dv200-130-indoor-cart/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..8225115bbe9a5b841eccdb2816a778541d448f9e
--- /dev/null
+++ b/images/verified_first_frame/dv200-130-indoor-cart/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d24d5c4b857c6d0362f478543087b2d12a1fbb252d04a0338b8ad7bb5b6f1ce4
+size 1019721
diff --git a/images/verified_first_frame/dv200-134-indoor-push/first-person.png b/images/verified_first_frame/dv200-134-indoor-push/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..6717c403814023f0e7db452e86d501f2b6968dc9
--- /dev/null
+++ b/images/verified_first_frame/dv200-134-indoor-push/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecf170f5cef297a2539a7ce47a5f12894287e7a757f09407c8fd03cae26d8c9f
+size 1105889
diff --git a/images/verified_first_frame/dv200-138-indoor-carryChair/follow.png b/images/verified_first_frame/dv200-138-indoor-carryChair/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..23356c77afed778376935d715ef8daab13698569
--- /dev/null
+++ b/images/verified_first_frame/dv200-138-indoor-carryChair/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee4c1b230246eed8229acf618d5a65b71e9f68625cc9b151cd05a3fc49efc77f
+size 783881
diff --git a/images/verified_first_frame/dv200-150-nature-dog-encounter/first-person.png b/images/verified_first_frame/dv200-150-nature-dog-encounter/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..66728252e9f8a0c80807f4baeeba3c283c7795fd
--- /dev/null
+++ b/images/verified_first_frame/dv200-150-nature-dog-encounter/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10a6d776d05a9f3863ef76754d6bbb5ebe12ac8a9704ad6a63c1b8eeae0c23a6
+size 826534
diff --git a/images/verified_first_frame/dv200-152-city-cat-encounter/follow.png b/images/verified_first_frame/dv200-152-city-cat-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..f6a5c552c2f41f6184b080100101210dc8cef5ad
--- /dev/null
+++ b/images/verified_first_frame/dv200-152-city-cat-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aec606bb9236e2e7225b13441f93497787de4157b54a0ce365e6ff3708e3d69
+size 799343
diff --git a/images/verified_first_frame/dv200-153-city-cat-encounter/follow.png b/images/verified_first_frame/dv200-153-city-cat-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..83e3a8340768b4765ed9316726ef3256e9beba14
--- /dev/null
+++ b/images/verified_first_frame/dv200-153-city-cat-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61316fe271129396c6d687792a7eab02d55ed97936fa4a5727f9af3851c67780
+size 872346
diff --git a/images/verified_first_frame/dv200-160-indoor-cat-encounter/first-person.png b/images/verified_first_frame/dv200-160-indoor-cat-encounter/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..bcb6749eefe1b83d895c33584cf7026ab04ba441
--- /dev/null
+++ b/images/verified_first_frame/dv200-160-indoor-cat-encounter/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c643d463a2a4a737744525f60c167a4bb4f50e04edebb505d5014f01591d14e5
+size 747505
diff --git a/images/verified_first_frame/dv200-164-nature-rabbit-encounter/follow.png b/images/verified_first_frame/dv200-164-nature-rabbit-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..ce7201c9626eaefdcd07dbbac06f95d1e83e5afc
--- /dev/null
+++ b/images/verified_first_frame/dv200-164-nature-rabbit-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b761bb684fa4be8fa4d37afd4654dbc93c97e71ac2805924389bb8bced754bf
+size 623030
diff --git a/images/verified_first_frame/dv200-168-city-rabbit-encounter/first-person.png b/images/verified_first_frame/dv200-168-city-rabbit-encounter/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..4f937ab3ee0c444bd53b9900b9cbe32fcf58e1d7
--- /dev/null
+++ b/images/verified_first_frame/dv200-168-city-rabbit-encounter/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:763587cded254502215d9e22ad4b1098dec2bec05413f20ea5d9921d6baf2dd2
+size 884116
diff --git a/images/verified_first_frame/dv200-174-nature-bird-encounter/follow.png b/images/verified_first_frame/dv200-174-nature-bird-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..22f6365d07f169a74638f25d94269ee1ed7f9ad9
--- /dev/null
+++ b/images/verified_first_frame/dv200-174-nature-bird-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c863a0b03079fe9a6a275e62272bfb7d3bb37ed0a65ce2ac1ff8221a3a2a408f
+size 1058381
diff --git a/images/verified_first_frame/dv200-181-city-duck-encounter/follow.png b/images/verified_first_frame/dv200-181-city-duck-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..194396cfb8eca5a7992ababd9ac9a4a831d088b1
--- /dev/null
+++ b/images/verified_first_frame/dv200-181-city-duck-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d98f94f5c1c055ff9425773d3bc194ac482119d7395aed9cdb74c7fc244911e0
+size 884760
diff --git a/images/verified_first_frame/dv200-184-city-duck-encounter/follow.png b/images/verified_first_frame/dv200-184-city-duck-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..0e84aaeff3eca03f2d5cbd39170a102fbd825ae4
--- /dev/null
+++ b/images/verified_first_frame/dv200-184-city-duck-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cf70bdb3d619d939c9bb0641fb2d2281be3ed173e9c8e2ef4f21aa588b2317b
+size 520737
diff --git a/images/verified_first_frame/dv200-186-city-duck-encounter/first-person.png b/images/verified_first_frame/dv200-186-city-duck-encounter/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..f9282ef8e19628c2d2a0d78a42fe144145a6a2c1
--- /dev/null
+++ b/images/verified_first_frame/dv200-186-city-duck-encounter/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cff0873ca4aea6c95feb1247c33c085b7f9b1a9f95ec54fc1e1ba679d746e9fa
+size 739829
diff --git a/images/verified_first_frame/dv200-187-nature-duck-encounter/first-person.png b/images/verified_first_frame/dv200-187-nature-duck-encounter/first-person.png
new file mode 100644
index 0000000000000000000000000000000000000000..b1176feb3333664ca00ff631e14731775caa479c
--- /dev/null
+++ b/images/verified_first_frame/dv200-187-nature-duck-encounter/first-person.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc5e814ffec3095231a431bfbfb2a7a31b1d79de63ff6db210433642508ff1e3
+size 869986
diff --git a/images/verified_first_frame/dv200-190-nature-duck-encounter/follow.png b/images/verified_first_frame/dv200-190-nature-duck-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..f26d3b8b138809b96db515eebf97d5459252eb92
--- /dev/null
+++ b/images/verified_first_frame/dv200-190-nature-duck-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5f257f7f0d1fc10331a099ffd9b2fda194dc2eedfc1fda2d34e50b316a53d02
+size 796004
diff --git a/images/verified_first_frame/dv200-199-underwater-fish-encounter/follow.png b/images/verified_first_frame/dv200-199-underwater-fish-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..9a259b94f3449a0d8a8e58c3953d98976fe3dab5
--- /dev/null
+++ b/images/verified_first_frame/dv200-199-underwater-fish-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e69d1e9fe698601f1b9a721b815116446e777db665bc219d37a8752f7352ee4
+size 815093
diff --git a/images/verified_first_frame/dv200-200-underwater-fish-encounter/follow.png b/images/verified_first_frame/dv200-200-underwater-fish-encounter/follow.png
new file mode 100644
index 0000000000000000000000000000000000000000..fd4515701f11016e361384df4e9a5193ff54ddcf
--- /dev/null
+++ b/images/verified_first_frame/dv200-200-underwater-fish-encounter/follow.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:684d2ea2c306cf9a17275cb6f6e05dd160b94bdf53f8ab54df3e08f0d503690b
+size 1042029
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b69718bbc353d2f72b87781cc6ad1cb13a5e692a
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout where green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail in two offset lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short dark hair, wearing a zip-up hiking jacket in royal blue, grey hiking trousers and brown leather boots. The other participants of the same recording are participant B, a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes; participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots and participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk one behind the other along one side of the trail while participants B and D walk one behind the other along the other side, C ahead of A and D ahead of B, the two lines offset so that nobody moves directly abreast; everyone keeps the same steady pace; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone keeps walking forward at the same steady pace, holding their places in the two offset lines with the spacing unchanged; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..25d1c0d7a47a72456d6186e3d1b80b5132ec6b22
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail in two offset lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short dark hair, wearing a zip-up hiking jacket in royal blue, grey hiking trousers and brown leather boots. The other participants of the same recording are participant B, a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes; participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots and participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk one behind the other along one side of the trail while participants B and D walk one behind the other along the other side, C ahead of A and D ahead of B, the two lines offset so that nobody moves directly abreast; everyone keeps the same steady pace; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone keeps walking forward at the same steady pace, holding their places in the two offset lines with the spacing unchanged; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/white/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..70480af61e35cae58be5cb4f41ff68a9fbaa583b
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail in two offset lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short dark hair, wearing a zip-up hiking jacket in royal blue, grey hiking trousers and brown leather boots. The other participants of the same recording are participant B, a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes; participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots and participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk one behind the other along one side of the trail while participants B and D walk one behind the other along the other side, C ahead of A and D ahead of B, the two lines offset so that nobody moves directly abreast; everyone keeps the same steady pace; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone keeps walking forward at the same steady pace, holding their places in the two offset lines with the spacing unchanged; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..22f7dead95350cf3f93de8191dd24b92eb136bdb
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant C), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout where green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes. The other people in this view are participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots and participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant C) is a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots, and keeps exactly this appearance. <Subject 4> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant D) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0f152e939b5303ff34e2be7db5b9d6e7fc1d39f7
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant C), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes. The other people in this view are participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots and participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant C) is a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots, and keeps exactly this appearance. <Subject 4> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant D) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/white/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6712d01f05c919f182d5597ffe6ffba0590dd52f
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-b/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant C), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with cropped light-brown hair, wearing a hooded sweatshirt in bright lime green, khaki cargo trousers and grey trail shoes. The other people in this view are participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots and participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant C) is a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots, and keeps exactly this appearance. <Subject 4> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant D) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..adb5a59dbd76d9c1ce559b3d09cd98233654ee22
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,34 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 2 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout where green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail, involving <Subject 3>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots. The other people in this view are participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6e86d07070e787ca1e1980ea1da6878ea65e59a0
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/mixed/prompt.txt
@@ -0,0 +1,34 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 2 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail, involving <Subject 3>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots. The other people in this view are participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/white/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..31d4b2fce44b849148689e7c0e20c982de46aeb7
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-c/white/prompt.txt
@@ -0,0 +1,34 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 2 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail, involving <Subject 3>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The people walk along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a slim adult with long dark hair in a ponytail, wearing a fleece pullover in vivid magenta pink, black leggings and dark hiking boots. The other people in this view are participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant D) is a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..857fc0160165d6a87ea7c300205b0b8a373bac65
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,32 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 1 human proxy standing for <Subject 1> (the protagonist, participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The protagonist walks along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout where green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The protagonist walks along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots. The protagonist keeps one consistent face, build and outfit for the whole clip and carries nothing. The protagonist keeps the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, the only subject in frame, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the protagonist moves steadily along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the protagonist keeps moving along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e3c048be62ae9e5df49837e2adf16d0c29071a06
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/mixed/prompt.txt
@@ -0,0 +1,32 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 1 blue mannequin proxy standing for <Subject 1> (the protagonist, participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The protagonist walks along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The protagonist walks along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots. The protagonist keeps one consistent face, build and outfit for the whole clip and carries nothing. The protagonist keeps the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, the only subject in frame, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the protagonist moves steadily along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the protagonist keeps moving along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/white/prompt.txt b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fd670e66e45a9b1abe4999ba6082f2247821d27a
--- /dev/null
+++ b/prompts/multi_view/multi4-001-forest-staggered-follow/follow-d/white/prompt.txt
@@ -0,0 +1,32 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 1 grey block-figure proxy standing for <Subject 1> (the protagonist, participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The protagonist walks along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a temperate forest trail. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, level earth trail through temperate woodland: tall broad-leaved trees with large trunks stand scattered on both sides, with fallen timber among them, and more trees recede along the trail under an open sky. The protagonist walks along the open trail. Preserve the reference spacing, tree positions and open walking routes. Use packed earth, leaf litter, rough bark, moss and green foliage. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a stocky adult with short greying hair and a trimmed beard, wearing a windbreaker in bright turquoise, beige trousers and dark brown boots. The protagonist keeps one consistent face, build and outfit for the whole clip and carries nothing. The protagonist keeps the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, the only subject in frame, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the protagonist moves steadily along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the protagonist keeps moving along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e57b48f290dc6bb1cfc12d08907d08aea24e8cc7
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout where blue-grey surfaces are water, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots. The other participants of the same recording are participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals; participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; the protagonist (participant A) moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. The protagonist (participant A) in the foreground covers participant B from about 3.6 s to 5.0 s; the continuous water surface in the foreground covers participant D from about 4.1 s to 4.6 s; the broad-leaf tropical tree in the foreground covers participant D from about 4.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4c60f8af877818358f73e10688f854ae665e344d
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots. The other participants of the same recording are participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals; participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; the protagonist (participant A) moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. The protagonist (participant A) in the foreground covers participant B from about 3.6 s to 5.0 s; the continuous water surface in the foreground covers participant D from about 4.1 s to 4.6 s; the broad-leaf tropical tree in the foreground covers participant D from about 4.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/white/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..dd6c72ce83bef03e00f14375b05d9a2799d2abb2
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots. The other participants of the same recording are participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals; participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; the protagonist (participant A) moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. The protagonist (participant A) in the foreground covers participant B from about 3.6 s to 5.0 s; the continuous water surface in the foreground covers participant D from about 4.1 s to 4.6 s; the broad-leaf tropical tree in the foreground covers participant D from about 4.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..22a84dc7dfd90046273cd441d773a19eb9eccea1
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout where blue-grey surfaces are water, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. Participant A in the foreground covers the protagonist from about 3.7 s to 5.0 s; the broad-leaf tropical tree in the foreground covers participant C from about 4.5 s to 4.9 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..04286981521664b770c61843c8510426e8b03c9c
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. Participant A in the foreground covers the protagonist from about 3.7 s to 5.0 s; the broad-leaf tropical tree in the foreground covers participant C from about 4.5 s to 4.9 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/white/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2046ba142076cdb6a81bf7e2d42907d86847d8cb
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. Participant A in the foreground covers the protagonist from about 3.7 s to 5.0 s; the broad-leaf tropical tree in the foreground covers participant C from about 4.5 s to 4.9 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6aaa99d69cb6e76e052f2c1dc8f9c33c169be2ad
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout where blue-grey surfaces are water, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the middle part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands in the centre of the middle part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. The continuous water surface in the foreground covers participant A from about 2.3 s to 3.2 s; the protagonist (participant C) in the foreground covers participant D from about 3.0 s to 5.0 s; the continuous water surface in the foreground covers participant D from about 3.4 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..87eae582564f4f894af4e853d615d150f459a4ea
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the middle part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands in the centre of the middle part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. The continuous water surface in the foreground covers participant A from about 2.3 s to 3.2 s; the protagonist (participant C) in the foreground covers participant D from about 3.0 s to 5.0 s; the continuous water surface in the foreground covers participant D from about 3.4 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/white/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..349cb5cfcc980d5ec3aced14db0e3425b92009db
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-c/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals and participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 5> (participant D) is a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the middle part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands in the centre of the middle part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route, drops about half a metre. The continuous water surface in the foreground covers participant A from about 2.3 s to 3.2 s; the protagonist (participant C) in the foreground covers participant D from about 3.0 s to 5.0 s; the continuous water surface in the foreground covers participant D from about 3.4 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e451f5a858393f77ab0ce32846b0efaf0d55a72d
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout where blue-grey surfaces are water, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals and participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 5> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 5>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the middle part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant C) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route, drops about half a metre. Participant C in the foreground covers participant A from about 0.0 s to 0.7 s; the continuous water surface in the foreground covers participant A from about 0.8 s to 2.7 s; the broad-leaf tropical tree in the foreground covers participant A from about 4.5 s to 4.8 s; the broad-leaf tropical tree in the foreground covers participant A from about 4.8 s to 5.0 s; the broad-leaf tropical tree in the foreground covers participant B from about 3.2 s to 3.8 s; participant C in the foreground covers the protagonist from about 3.0 s to 5.0 s; the continuous water surface in the foreground covers the protagonist from about 3.0 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3adbc0f166e8d9a473043cc862d8e1fe0c5cfcc5
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals and participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 5> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 5>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the middle part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant C) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route, drops about half a metre. Participant C in the foreground covers participant A from about 0.0 s to 0.7 s; the continuous water surface in the foreground covers participant A from about 0.8 s to 2.7 s; the broad-leaf tropical tree in the foreground covers participant A from about 4.5 s to 4.8 s; the broad-leaf tropical tree in the foreground covers participant A from about 4.8 s to 5.0 s; the broad-leaf tropical tree in the foreground covers participant B from about 3.2 s to 3.8 s; participant C in the foreground covers the protagonist from about 3.0 s to 5.0 s; the continuous water surface in the foreground covers the protagonist from about 3.0 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/white/prompt.txt b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..10cfca5a9873f235ccd127ca021c59158962cbc6
--- /dev/null
+++ b/prompts/multi_view/multi4-002-jungle-opposing-pairs/follow-d/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> wading and walking through a tropical jungle clearing as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A clearing in dense tropical jungle crossed by a shallow stream: tall thin-trunked trees with broad leafy crowns and hanging vines surround a level stretch of muddy ground, and the clear, shin-deep stream runs across one end of it. The people wade through the stream and walk on the open ground. Preserve the reference spacing, tree positions, the stream edge and open walking routes. Use wet mud, shallow clear water, smooth bark, broad glossy leaves and hanging vines. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a tall slim adult with a short brown bob, wearing a long-sleeved sun shirt in bright turquoise, dark grey trousers and brown rubber boots. The other participants of the same recording are participant A, a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots; participant B, a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals and participant C, an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a lean adult with short black hair, wearing a lightweight rain jacket in royal blue, rolled-up grey trousers and black rubber boots, and keeps exactly this appearance. <Subject 4> (participant B) is a broad-shouldered adult with a shaved head, wearing a short-sleeved field shirt in bright lime green, khaki shorts and brown sandals, and keeps exactly this appearance. <Subject 5> (participant C) is an adult of average build with shoulder-length curly hair, wearing a sleeveless top in vivid magenta pink, beige cargo shorts and black water shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 5>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the middle part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant C) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B wade out of the shallow stream one behind the other, B ahead of A, and walk on along one lane while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route, rises about half a metre; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - while D steps down into the shallow stream as the pairs move apart; participant A moves about 3 metres along its route, rises about half a metre; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route, drops about half a metre. Participant C in the foreground covers participant A from about 0.0 s to 0.7 s; the continuous water surface in the foreground covers participant A from about 0.8 s to 2.7 s; the broad-leaf tropical tree in the foreground covers participant A from about 4.5 s to 4.8 s; the broad-leaf tropical tree in the foreground covers participant A from about 4.8 s to 5.0 s; the broad-leaf tropical tree in the foreground covers participant B from about 3.2 s to 3.8 s; participant C in the foreground covers the protagonist from about 3.0 s to 5.0 s; the continuous water surface in the foreground covers the protagonist from about 3.0 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5b9c2539a5909fd61d506845ed8f9514d4d3b9ff
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots. The other participants of the same recording are participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots; participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. <Subject 5> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the side, facing the screen-right; <Subject 5> (participant D) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward an open patch of pasture along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant D in the foreground covers the protagonist from about 3.9 s to 4.2 s; participant D in the foreground covers participant B from about 2.7 s to 3.6 s; the protagonist (participant A) in the foreground covers participant C from about 0.9 s to 2.3 s; participant D in the foreground covers participant C from about 3.3 s to 3.8 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4345726aa5450e948ff6ca698cf0f303f85cb21f
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots. The other participants of the same recording are participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots; participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. <Subject 5> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the side, facing the screen-right; <Subject 5> (participant D) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward an open patch of pasture along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant D in the foreground covers the protagonist from about 3.9 s to 4.2 s; participant D in the foreground covers participant B from about 2.7 s to 3.6 s; the protagonist (participant A) in the foreground covers participant C from about 0.9 s to 2.3 s; participant D in the foreground covers participant C from about 3.3 s to 3.8 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/white/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a6f1ba298dfc037e3c46b5931e42128d4991050f
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots. The other participants of the same recording are participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots; participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. <Subject 5> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the side, facing the screen-right; <Subject 5> (participant D) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward an open patch of pasture along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant D in the foreground covers the protagonist from about 3.9 s to 4.2 s; participant D in the foreground covers participant B from about 2.7 s to 3.6 s; the protagonist (participant A) in the foreground covers participant C from about 0.9 s to 2.3 s; participant D in the foreground covers participant C from about 3.3 s to 3.8 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..93120ca078b93c576b164ffbf1bc994bf90f8007
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots. The other participants of the same recording are participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots; participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. <Subject 5> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from the side, facing the screen-right. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward an open patch of pasture along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant C in the foreground covers participant A from about 2.7 s to 3.4 s; participant C in the foreground covers the protagonist from about 3.8 s to 4.2 s; the protagonist (participant B) in the foreground covers participant D from about 0.9 s to 2.3 s; participant C in the foreground covers participant D from about 3.2 s to 3.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..75afb3041662589128093cbbdf99200ff480e159
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots. The other participants of the same recording are participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots; participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. <Subject 5> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from the side, facing the screen-right. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward an open patch of pasture along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant C in the foreground covers participant A from about 2.7 s to 3.4 s; participant C in the foreground covers the protagonist from about 3.8 s to 4.2 s; the protagonist (participant B) in the foreground covers participant D from about 0.9 s to 2.3 s; participant C in the foreground covers participant D from about 3.2 s to 3.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/white/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a876278b57f4829d47ad0b22ffa559441452376b
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots. The other participants of the same recording are participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots; participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. <Subject 5> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from the side, facing the screen-right. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward an open patch of pasture along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant C in the foreground covers participant A from about 2.7 s to 3.4 s; participant C in the foreground covers the protagonist from about 3.8 s to 4.2 s; the protagonist (participant B) in the foreground covers participant D from about 0.9 s to 2.3 s; participant C in the foreground covers participant D from about 3.2 s to 3.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7545c4b9436f66ea7d9a8c88b4c1afbe7ee264a0
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant B), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots. The other people in this view are participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots, and keeps exactly this appearance. <Subject 4> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. The protagonist (participant C) in the foreground covers participant B from about 1.3 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..60f5046e79a27fb8284b515649a3f3a54ec225b0
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant B), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots. The other people in this view are participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots, and keeps exactly this appearance. <Subject 4> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. The protagonist (participant C) in the foreground covers participant B from about 1.3 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/white/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..26e1ad8d31f4bab8f22dece34cf2153c32ca5057
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-c/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant B), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots. The other people in this view are participant B, a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots and participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a slim adult with a long dark braid, wearing a knitted cardigan in bright lime green, a white T-shirt, beige trousers and tan ankle boots, and keeps exactly this appearance. <Subject 4> (participant D) is an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. The protagonist (participant C) in the foreground covers participant B from about 1.3 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0b61fda87f53932acb67f4ccd8c03b54726b741a
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots. The other people in this view are participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots and participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant C) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The protagonist (participant D) in the foreground covers participant A from about 1.1 s to 2.2 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..02362a3c4e39feb93136004af69865d3f898ebd6
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots. The other people in this view are participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots and participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant C) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The protagonist (participant D) in the foreground covers participant A from about 1.1 s to 2.2 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/white/prompt.txt b/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c60f809c036605cff6f5f41fac90c210649fc219
--- /dev/null
+++ b/prompts/multi_view/multi4-003-grassland-crossroads/follow-d/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across open grassland, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: Open, level grassland with fenced paddocks: long wooden rail fences run across the pasture, a large barn stands to one side, scattered trees dot the rolling ground and the grass stretches away to the horizon under an open sky. The people walk on the open grass between the fences. Preserve the reference spacing, fence lines, barn and tree positions and open walking routes. Use short green grass, weathered wooden rails, painted barn boards and leafy trees. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, an adult of medium build with closely cropped grey hair, wearing a cotton work shirt in bright turquoise, khaki trousers and brown leather boots. The other people in this view are participant A, a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots and participant C, a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short sandy hair, wearing a plaid flannel shirt in royal blue and white, faded grey jeans and brown work boots, and keeps exactly this appearance. <Subject 4> (participant C) is a tall adult with short wavy auburn hair, wearing a quilted gilet in vivid magenta pink over a grey shirt, black trousers and dark rubber boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant C) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The protagonist (participant D) in the foreground covers participant A from about 1.1 s to 2.2 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..28b3a20054c2dd77bdfc12d8a462f91049859f4a
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout where grey-brown shapes are rocks, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots. The other participants of the same recording are participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots; participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. <Subject 5> (participant D) is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; the protagonist (participant A) moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; the protagonist (participant A) moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. The protagonist (participant A) in the foreground covers participant C from about 2.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ce4f3a3afce683b4bac0594780e452be64493aad
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots. The other participants of the same recording are participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots; participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. <Subject 5> (participant D) is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; the protagonist (participant A) moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; the protagonist (participant A) moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. The protagonist (participant A) in the foreground covers participant C from about 2.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/white/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1bbfe4bc8a9e65bb26634404c9ab07ee9327e517
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots. The other participants of the same recording are participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots; participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. <Subject 5> (participant D) is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; the protagonist (participant A) moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; the protagonist (participant A) moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. The protagonist (participant A) in the foreground covers participant C from about 2.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..14e1bdb89408c3e5df757116ba8095e7591a0181
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout where grey-brown shapes are rocks, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7 that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots. The other participants of the same recording are participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots; participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. <Subject 5> (participant D) is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands at the far screen-left edge of the lower part of the frame, filling most of the frame height, nearest to the camera, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; participant A moves about 3 metres along its route, drops a short distance; the protagonist (participant B) moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; participant A moves about 3 metres along its route, drops a short distance; the protagonist (participant B) moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. The protagonist (participant B) in the foreground covers participant D from about 2.1 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..dea80540ecb9069a5e40cacf3021cbd64e36ea7b
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots. The other participants of the same recording are participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots; participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. <Subject 5> (participant D) is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands at the far screen-left edge of the lower part of the frame, filling most of the frame height, nearest to the camera, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; participant A moves about 3 metres along its route, drops a short distance; the protagonist (participant B) moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; participant A moves about 3 metres along its route, drops a short distance; the protagonist (participant B) moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. The protagonist (participant B) in the foreground covers participant D from about 2.1 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/white/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8527dc9d910a59e320e0b0943cba2e1c3a3f2cae
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots. The other participants of the same recording are participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots; participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. <Subject 5> (participant D) is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands at the far screen-left edge of the lower part of the frame, filling most of the frame height, nearest to the camera, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; participant A moves about 3 metres along its route, drops a short distance; the protagonist (participant B) moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; participant A moves about 3 metres along its route, drops a short distance; the protagonist (participant B) moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. The protagonist (participant B) in the foreground covers participant D from about 2.1 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..013429188cdc732cbdf4e65e3970ba989ab4eebf
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 1.5 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout where grey-brown shapes are rocks, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D: not in frame at 00:00.000; it enters the frame at about 1.5 s; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (enters [Shot 1] at about 1.5 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots. The other participants of the same recording are participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots; participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 5> (participant D) is not in frame at 00:00.000 and enters the frame at about 1.5 s; from then on it is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the middle part of the frame, filling most of the frame height, nearest to the camera, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant B) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; participant A moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; the protagonist (participant C) moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; participant A moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; the protagonist (participant C) moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. Participant A in the foreground covers the protagonist from about 1.5 s to 5.0 s; participant A in the foreground covers participant D from about 0.0 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..83c81202f0402fd92c242c51155b382574a03c12
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 1.5 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D: not in frame at 00:00.000; it enters the frame at about 1.5 s; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (enters [Shot 1] at about 1.5 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots. The other participants of the same recording are participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots; participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 5> (participant D) is not in frame at 00:00.000 and enters the frame at about 1.5 s; from then on it is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the middle part of the frame, filling most of the frame height, nearest to the camera, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant B) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; participant A moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; the protagonist (participant C) moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; participant A moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; the protagonist (participant C) moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. Participant A in the foreground covers the protagonist from about 1.5 s to 5.0 s; participant A in the foreground covers participant D from about 0.0 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/white/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5fc62a9d0d005ccffa3cd1f9f8c288f9e7c26619
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-c/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 1.5 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D: not in frame at 00:00.000; it enters the frame at about 1.5 s; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope side by side in two parallel lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (enters [Shot 1] at about 1.5 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots. The other participants of the same recording are participant A, an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots; participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots and participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short dark hair, wearing a padded ski jacket in royal blue, black snow trousers, grey gloves and black snow boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 5> (participant D) is not in frame at 00:00.000 and enters the frame at about 1.5 s; from then on it is a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the middle part of the frame, filling most of the frame height, nearest to the camera, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant B) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk downhill forward in two parallel lines along the snow slope, C ahead of A on one side and D ahead of B on the other, all angled slightly the same way at the same steady pace; participant A moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; the protagonist (participant C) moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, together they ease out of the slight turn and keep walking downhill straight ahead in parallel, holding their spacing; participant A moves about 3 metres along its route, drops a short distance; participant B moves about 3 metres along its route, drops a short distance; the protagonist (participant C) moves about 3 metres along its route, drops a short distance; participant D moves about 3 metres along its route, drops a short distance. Participant A in the foreground covers the protagonist from about 1.5 s to 5.0 s; participant A in the foreground covers participant D from about 0.0 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..158a78b9a48f99587a4bb012b4f05e658bec32c6
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout where grey-brown shapes are rocks, green low-poly shapes are trees, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots. The other people in this view are participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots and participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, filling most of the frame height, nearest to the camera, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands at the far screen-left edge of the lower part of the frame, filling most of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; the protagonist (participant D) moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; the protagonist (participant D) moves about 3 metres along its route, drops a short distance. Participant B in the foreground covers the protagonist from about 0.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..dc489dafcce93672ab434c6cc7caeb34fe002545
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots. The other people in this view are participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots and participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, filling most of the frame height, nearest to the camera, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands at the far screen-left edge of the lower part of the frame, filling most of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; the protagonist (participant D) moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; the protagonist (participant D) moves about 3 metres along its route, drops a short distance. Participant B in the foreground covers the protagonist from about 0.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/white/prompt.txt b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..359e5ee3610c441dcd93e42dcb74065e20152b98
--- /dev/null
+++ b/prompts/multi_view/multi4-006-snow-parallel-turn/follow-d/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking down a gentle snow slope, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A wide, gently falling snowfield: dark conifers stand in groups on both sides, faceted ice and dark rock formations break through the snow here and there, and a frozen open path runs down the middle under a pale sky. The people walk down the open snow. Preserve the reference spacing, conifer, rock and ice positions, the falling ground and open walking routes. Use fresh snow, packed snow, clear blue-white ice, dark rock and deep green conifer needles. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a broad adult with short black curly hair, wearing an insulated anorak in bright turquoise, charcoal trousers, black mittens and dark brown boots. The other people in this view are participant B, a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots and participant C, a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown beard, wearing a hooded parka in bright lime green, grey snow trousers, a black beanie and brown winter boots, and keeps exactly this appearance. <Subject 4> (participant C) is a slim adult with long copper hair, wearing a puffer jacket in vivid magenta pink, white snow trousers, a grey knitted hat and white snow boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, filling most of the frame height, nearest to the camera, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands at the far screen-left edge of the lower part of the frame, filling most of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; the protagonist (participant D) moves about 3 metres along its route, drops a short distance. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route, drops a short distance; participant C moves about 3 metres along its route, drops a short distance; the protagonist (participant D) moves about 3 metres along its route, drops a short distance. Participant B in the foreground covers the protagonist from about 0.8 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fb1f67ac463cca60e5ecec145a05de8ea75372b7
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout where grey-brown shapes are rocks, light-green low shapes are ground vegetation, blue-grey surfaces are water, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins. The other participants of the same recording are participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins; participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 4> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 5> (participant D) is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure close by, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure close by, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, tiny in the distance, in the far background, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. The protagonist (participant A) in the foreground covers participant C from about 0.1 s to 0.5 s; the protagonist (participant A) in the foreground covers participant C from about 0.7 s to 1.1 s; the protagonist (participant A) in the foreground covers participant C from about 4.6 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d9a4682af217336cb0b3e1c12413e0c650cdcac4
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins. The other participants of the same recording are participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins; participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 4> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 5> (participant D) is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure close by, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure close by, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, tiny in the distance, in the far background, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. The protagonist (participant A) in the foreground covers participant C from about 0.1 s to 0.5 s; the protagonist (participant A) in the foreground covers participant C from about 0.7 s to 1.1 s; the protagonist (participant A) in the foreground covers participant C from about 4.6 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/white/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..80645cfcaf5fa534de4d75073736fbee279cd6c7
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins. The other participants of the same recording are participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins; participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 4> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 5> (participant D) is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure close by, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure close by, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, tiny in the distance, in the far background, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. The protagonist (participant A) in the foreground covers participant C from about 0.1 s to 0.5 s; the protagonist (participant A) in the foreground covers participant C from about 0.7 s to 1.1 s; the protagonist (participant A) in the foreground covers participant C from about 4.6 s to 5.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3b6eb928c57e0b33c0f1b186d9bb14a5a29a4726
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 0.3 s.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout where grey-brown shapes are rocks, light-green low shapes are ground vegetation, blue-grey surfaces are water, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7 that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D: not in frame at 00:00.000; it enters the frame at about 0.3 s; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (enters [Shot 1] at about 0.3 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 5> (participant D) is not in frame at 00:00.000 and enters the frame at about 0.3 s; from then on it is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands at the far screen-left edge of the middle part of the frame, a small figure, nearest to the camera, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure in the middle distance, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. The protagonist (participant B) in the foreground covers participant D from about 0.0 s to 0.4 s; the protagonist (participant B) in the foreground covers participant D from about 0.5 s to 1.0 s; the protagonist (participant B) in the foreground covers participant D from about 1.2 s to 1.6 s; the protagonist (participant B) in the foreground covers participant D from about 1.9 s to 2.1 s; the protagonist (participant B) in the foreground covers participant D from about 2.5 s to 2.8 s; the protagonist (participant B) in the foreground covers participant D from about 3.1 s to 3.5 s; the protagonist (participant B) in the foreground covers participant D from about 3.7 s to 4.1 s; the protagonist (participant B) in the foreground covers participant D from about 4.3 s to 4.8 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f755c842e187ded98eaf3d3bf1d73360420cfd24
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 0.3 s.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D: not in frame at 00:00.000; it enters the frame at about 0.3 s; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (enters [Shot 1] at about 0.3 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 5> (participant D) is not in frame at 00:00.000 and enters the frame at about 0.3 s; from then on it is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands at the far screen-left edge of the middle part of the frame, a small figure, nearest to the camera, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure in the middle distance, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. The protagonist (participant B) in the foreground covers participant D from about 0.0 s to 0.4 s; the protagonist (participant B) in the foreground covers participant D from about 0.5 s to 1.0 s; the protagonist (participant B) in the foreground covers participant D from about 1.2 s to 1.6 s; the protagonist (participant B) in the foreground covers participant D from about 1.9 s to 2.1 s; the protagonist (participant B) in the foreground covers participant D from about 2.5 s to 2.8 s; the protagonist (participant B) in the foreground covers participant D from about 3.1 s to 3.5 s; the protagonist (participant B) in the foreground covers participant D from about 3.7 s to 4.1 s; the protagonist (participant B) in the foreground covers participant D from about 4.3 s to 4.8 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/white/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0e1f0e4c0a0fecc232aaea56e319ac522d63653d
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 0.3 s.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D: not in frame at 00:00.000; it enters the frame at about 0.3 s; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (enters [Shot 1] at about 0.3 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 5> (participant D) is not in frame at 00:00.000 and enters the frame at about 0.3 s; from then on it is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands at the far screen-left edge of the middle part of the frame, a small figure, nearest to the camera, to the screen-left of <Subject 1>, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure in the middle distance, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 2 metres along its route; participant C moves about 2 metres along its route; participant D moves about 2 metres along its route. The protagonist (participant B) in the foreground covers participant D from about 0.0 s to 0.4 s; the protagonist (participant B) in the foreground covers participant D from about 0.5 s to 1.0 s; the protagonist (participant B) in the foreground covers participant D from about 1.2 s to 1.6 s; the protagonist (participant B) in the foreground covers participant D from about 1.9 s to 2.1 s; the protagonist (participant B) in the foreground covers participant D from about 2.5 s to 2.8 s; the protagonist (participant B) in the foreground covers participant D from about 3.1 s to 3.5 s; the protagonist (participant B) in the foreground covers participant D from about 3.7 s to 4.1 s; the protagonist (participant B) in the foreground covers participant D from about 4.3 s to 4.8 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4b4d1652b06003423e50f55120316f8cca078c4f
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout where grey-brown shapes are rocks, light-green low shapes are ground vegetation, blue-grey surfaces are water, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 5> (participant D) is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 2 metres along its route. Participant A in the foreground covers participant B from about 1.4 s to 1.9 s; participant A in the foreground covers participant B from about 1.9 s to 2.2 s; participant A in the foreground covers participant B from about 2.9 s to 3.1 s; participant A in the foreground covers participant B from about 3.2 s to 4.3 s; participant A in the foreground covers the protagonist from about 4.7 s to 4.9 s; participant A in the foreground covers participant D from about 0.8 s to 1.1 s; participant A in the foreground covers participant D from about 1.4 s to 1.7 s; participant A in the foreground covers participant D from about 2.0 s to 2.4 s; participant A in the foreground covers participant D from about 2.6 s to 3.0 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..207b844df4d9f8ee8135269e576855101fe5d983
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 5> (participant D) is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 2 metres along its route. Participant A in the foreground covers participant B from about 1.4 s to 1.9 s; participant A in the foreground covers participant B from about 1.9 s to 2.2 s; participant A in the foreground covers participant B from about 2.9 s to 3.1 s; participant A in the foreground covers participant B from about 3.2 s to 4.3 s; participant A in the foreground covers the protagonist from about 4.7 s to 4.9 s; participant A in the foreground covers participant D from about 0.8 s to 1.1 s; participant A in the foreground covers participant D from about 1.4 s to 1.7 s; participant A in the foreground covers participant D from about 2.0 s to 2.4 s; participant A in the foreground covers participant D from about 2.6 s to 3.0 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/white/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7bcbfa611aec396862352018164deabd202f0b7b
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-c/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins and participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 5> (participant D) is a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 2 metres along its route. Participant A in the foreground covers participant B from about 1.4 s to 1.9 s; participant A in the foreground covers participant B from about 1.9 s to 2.2 s; participant A in the foreground covers participant B from about 2.9 s to 3.1 s; participant A in the foreground covers participant B from about 3.2 s to 4.3 s; participant A in the foreground covers the protagonist from about 4.7 s to 4.9 s; participant A in the foreground covers participant D from about 0.8 s to 1.1 s; participant A in the foreground covers participant D from about 1.4 s to 1.7 s; participant A in the foreground covers participant D from about 2.0 s to 2.4 s; participant A in the foreground covers participant D from about 2.6 s to 3.0 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7b49a3df919e0e91aa00588e30b129e8589bd49b
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout where grey-brown shapes are rocks, light-green low shapes are ground vegetation, blue-grey surfaces are water, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7 that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins and participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 5> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, tiny in the distance, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 4>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant B) stands in the centre of the middle part of the frame, a small figure in the middle distance, nearest to the camera, seen from behind, facing away from the camera; <Subject 5> (participant C) stands on the screen-left of the middle part of the frame, tiny in the distance, in the far background, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 2 metres along its route. Participant B in the foreground covers participant C from about 1.3 s to 1.6 s; participant B in the foreground covers participant C from about 1.9 s to 2.3 s; participant B in the foreground covers participant C from about 2.5 s to 2.9 s; participant B in the foreground covers the protagonist from about 3.3 s to 3.5 s; participant B in the foreground covers the protagonist from about 3.7 s to 4.1 s; participant B in the foreground covers the protagonist from about 4.3 s to 4.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..497207e6926a299c20d519b5784582470cdef631
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins and participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 5> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, tiny in the distance, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 4>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant B) stands in the centre of the middle part of the frame, a small figure in the middle distance, nearest to the camera, seen from behind, facing away from the camera; <Subject 5> (participant C) stands on the screen-left of the middle part of the frame, tiny in the distance, in the far background, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 2 metres along its route. Participant B in the foreground covers participant C from about 1.3 s to 1.6 s; participant B in the foreground covers participant C from about 1.9 s to 2.3 s; participant B in the foreground covers participant C from about 2.5 s to 2.9 s; participant B in the foreground covers the protagonist from about 3.3 s to 3.5 s; participant B in the foreground covers the protagonist from about 3.7 s to 4.1 s; participant B in the foreground covers the protagonist from about 4.3 s to 4.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/white/prompt.txt b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d6cf6ab2eb2cf1d8aba9bb85f91059c6be9c6be9
--- /dev/null
+++ b/prompts/multi_view/multi4-010-underwater-converge-turn/follow-d/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> swimming underwater over a coral reef as two lines of divers angle toward each other, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An underwater reef seen below the surface: a sandy seabed with scattered dark rocks, clumps of branching coral and swaying sea plants, the rippling water surface overhead letting down soft light, and the blue water fading into haze in the distance. The people swim horizontally in open water above the seabed. Preserve the reference spacing, rock and coral positions and open swimming lanes. Use pale sand, dark volcanic rock, branching coral, green sea grass and clear blue water with light caustics. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a broad-shouldered adult diver in a full wetsuit in bright turquoise with black shoulders, a dark dive mask and white fins. The other participants of the same recording are participant A, a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins; participant B, a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins and participant C, a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a slim adult diver in a full wetsuit in royal blue with black side panels, a clear dive mask and black fins, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult diver in a full wetsuit in bright lime green with black sleeves, a black dive mask and grey fins, and keeps exactly this appearance. <Subject 5> (participant C) is a compact adult diver with long hair floating loose, in a full wetsuit in vivid magenta pink with grey panels, a clear dive mask and black fins, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, tiny in the distance, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 4>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant B) stands in the centre of the middle part of the frame, a small figure in the middle distance, nearest to the camera, seen from behind, facing away from the camera; <Subject 5> (participant C) stands on the screen-left of the middle part of the frame, tiny in the distance, in the far background, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C swim forward one behind the other on one side of the reef while participants B and D swim forward one behind the other on the other side, C ahead of A and D ahead of B, the two lines angling gently toward each other; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 2 metres along its route. From 00:02.500 to 00:05.000, a and C keep turning gently inward while B and D straighten out, so the two lines end closer together, still with clear space between them, everyone swimming at the same steady pace; participant A moves about 2 metres along its route; participant B moves about 2 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 2 metres along its route. Participant B in the foreground covers participant C from about 1.3 s to 1.6 s; participant B in the foreground covers participant C from about 1.9 s to 2.3 s; participant B in the foreground covers participant C from about 2.5 s to 2.9 s; participant B in the foreground covers the protagonist from about 3.3 s to 3.5 s; participant B in the foreground covers the protagonist from about 3.7 s to 4.1 s; participant B in the foreground covers the protagonist from about 4.3 s to 4.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4e576c7e21ad20119b53f78f749fe814f4655c30
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street in two offset lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short grey hair, wearing a cardigan in royal blue over a white shirt, dark brown trousers and brown shoes. The other participants of the same recording are participant B, a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers; participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats and participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers, and keeps exactly this appearance. <Subject 4> (participant C) is a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats, and keeps exactly this appearance. <Subject 5> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk one behind the other along one side of the market lane while participants B and D walk one behind the other along the other side, C ahead of A and D ahead of B, the two lines offset so that nobody moves directly abreast; everyone keeps the same steady pace; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone keeps walking forward at the same steady pace, holding their places in the two offset lines with the spacing unchanged; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..aa0b742d87287cb2028a4e8b5174070bdc7c81af
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street in two offset lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short grey hair, wearing a cardigan in royal blue over a white shirt, dark brown trousers and brown shoes. The other participants of the same recording are participant B, a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers; participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats and participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers, and keeps exactly this appearance. <Subject 4> (participant C) is a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats, and keeps exactly this appearance. <Subject 5> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk one behind the other along one side of the market lane while participants B and D walk one behind the other along the other side, C ahead of A and D ahead of B, the two lines offset so that nobody moves directly abreast; everyone keeps the same steady pace; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone keeps walking forward at the same steady pace, holding their places in the two offset lines with the spacing unchanged; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/white/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..35504ac6fde4df49d81da5d8dc1620d1902074e6
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street in two offset lines, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short grey hair, wearing a cardigan in royal blue over a white shirt, dark brown trousers and brown shoes. The other participants of the same recording are participant B, a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers; participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats and participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers, and keeps exactly this appearance. <Subject 4> (participant C) is a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats, and keeps exactly this appearance. <Subject 5> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk one behind the other along one side of the market lane while participants B and D walk one behind the other along the other side, C ahead of A and D ahead of B, the two lines offset so that nobody moves directly abreast; everyone keeps the same steady pace; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone keeps walking forward at the same steady pace, holding their places in the two offset lines with the spacing unchanged; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f6fa2f226e92ae8bff5fe4036ef45012e7279bf9
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant C), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers. The other people in this view are participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats and participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant C) is a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats, and keeps exactly this appearance. <Subject 4> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant D) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9d10c232918916ae8179b0b7109298b13ab31a0f
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant C), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers. The other people in this view are participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats and participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant C) is a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats, and keeps exactly this appearance. <Subject 4> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant D) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/white/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..dde159de2ad3c480c770152d654860fef3245172
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-b/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant C), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with short spiky black hair, wearing an oversized T-shirt in bright lime green, black cargo trousers and white trainers. The other people in this view are participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats and participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant C) is a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats, and keeps exactly this appearance. <Subject 4> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant D) stands in the centre of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..39b75d7e95d258aa1b9244971247931437eb55d5
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,34 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 2 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street, involving <Subject 3>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats. The other people in this view are participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1a1e549e683a6dc26d3b4baad2f93cd2a0fec46f
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/mixed/prompt.txt
@@ -0,0 +1,34 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 2 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street, involving <Subject 3>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats. The other people in this view are participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/white/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0ac1ccc89b3e3e550927127d83873343d2ad9faf
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-c/white/prompt.txt
@@ -0,0 +1,34 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 2 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street, involving <Subject 3>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The people walk along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a tall slim adult with long straight dark hair, wearing a wrap blouse in vivid magenta pink, cream trousers and black flats. The other people in this view are participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant D) is a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0196799bc6a5adf87560a65e47cb8f9200965f95
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,32 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 1 human proxy standing for <Subject 1> (the protagonist, participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The protagonist walks along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The protagonist walks along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals. The protagonist keeps one consistent face, build and outfit for the whole clip and carries nothing. The protagonist keeps the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, the only subject in frame, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the protagonist moves steadily along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the protagonist keeps moving along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6817cfd35550b0e2612cacc4900d4dbd5d515470
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/mixed/prompt.txt
@@ -0,0 +1,32 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 1 blue mannequin proxy standing for <Subject 1> (the protagonist, participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The protagonist walks along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The protagonist walks along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals. The protagonist keeps one consistent face, build and outfit for the whole clip and carries nothing. The protagonist keeps the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, the only subject in frame, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the protagonist moves steadily along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the protagonist keeps moving along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/white/prompt.txt b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4e6c1e65756ab4bb46e84b4c0937ec4241fec503
--- /dev/null
+++ b/prompts/multi_view/multi4-013-old-town-staggered-follow/follow-d/white/prompt.txt
@@ -0,0 +1,32 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 1 grey block-figure proxy standing for <Subject 1> (the protagonist, participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The protagonist walks along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through a narrow old-town market street. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: A narrow old-town market street: old stone buildings stand close on both sides, wooden market stalls with fabric awnings and a few handcarts sit along the edges, and the paved lane runs on between more facades under an open sky. The protagonist walks along the open middle of the lane. Preserve the reference spacing, building masses, stall positions and open walking routes. Use worn limestone, cobbles, weathered timber and faded canvas. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a heavy-set adult with a bald head and a grey moustache, wearing a button-up shirt in bright turquoise, beige trousers and dark leather sandals. The protagonist keeps one consistent face, build and outfit for the whole clip and carries nothing. The protagonist keeps the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, the only subject in frame, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the protagonist moves steadily along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the protagonist keeps moving along its route exactly as recorded in <Video 1>; the protagonist (participant D) moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..493558f6b755b463e7e152fbff5e9d238f7bd0db
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers. The other participants of the same recording are participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks; participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..919e998e5fd2fc17886b473a6109b91f2f0223ee
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers. The other participants of the same recording are participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks; participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-a/white/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..af6a970217cc4714a23410de4feff9a039c526d9
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers. The other participants of the same recording are participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks; participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..043b1e47987621cec66a4414a346decdffab54fb
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7 that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks. The other participants of the same recording are participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers; participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, nearest to the camera, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c60b62cf28a58f084fe8f7df7ccbf0d6dd8ebc20
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks. The other participants of the same recording are participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers; participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, nearest to the camera, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-b/white/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d336263da775971a1bb78f11d0a436363e621691
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks. The other participants of the same recording are participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers; participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, nearest to the camera, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..eb034835912f111fce6e0c605011c6c401d6d3e8
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 4> is not in frame at 00:00.000 and enters at about 0.7 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B: not in frame at 00:00.000; it enters the frame at about 0.7 s; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (enters [Shot 1] at about 0.7 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers. The other participants of the same recording are participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers; participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers, and keeps exactly this appearance. <Subject 4> (participant B) is not in frame at 00:00.000 and enters the frame at about 0.7 s; from then on it is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..daf8f02cabfba7f0b26ee620e3c60992966d286e
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-c/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 4> is not in frame at 00:00.000 and enters at about 0.7 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B: not in frame at 00:00.000; it enters the frame at about 0.7 s; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (enters [Shot 1] at about 0.7 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers. The other participants of the same recording are participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers; participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers, and keeps exactly this appearance. <Subject 4> (participant B) is not in frame at 00:00.000 and enters the frame at about 0.7 s; from then on it is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-c/white/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..27d449c63a95b8b5326a3ed04ad61898521e2009
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-c/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 4> is not in frame at 00:00.000 and enters at about 0.7 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B: not in frame at 00:00.000; it enters the frame at about 0.7 s; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (enters [Shot 1] at about 0.7 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers. The other participants of the same recording are participant A, an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers; participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks and participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an older adult with short grey hair, wearing a cardigan in royal blue over a white T-shirt, beige trousers and soft grey slippers, and keeps exactly this appearance. <Subject 4> (participant B) is not in frame at 00:00.000 and enters the frame at about 0.7 s; from then on it is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 5> (participant D) is a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk forward slowly one behind the other along the rooms while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..92df852180d767c65dbac28610debd1632970d01
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers. The other people in this view are participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks and participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..77b63f09ac24c4fac057193efdd281c67c908bb5
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-d/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers. The other people in this view are participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks and participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-017-home-overtake/follow-d/white/prompt.txt b/prompts/multi_view/multi4-017-home-overtake/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..02cf44be5d99318abd5e3fd87f360da590dac9d3
--- /dev/null
+++ b/prompts/multi_view/multi4-017-home-overtake/follow-d/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking through the connected rooms of a furnished home, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The interior of a roofed family home: connected rooms open into one another along a wide, straight route, with cabinets, shelves, chairs and dining tables standing along the walls, a flat ceiling overhead and warm indoor light. The people walk along the open floor between the furniture. Preserve the reference spacing, wall, cabinet, table and chair positions and open walking routes. Use polished wooden floorboards, painted plaster walls, varnished wood furniture and soft fabric. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a tall adult with a short black beard, wearing a sweatshirt in bright turquoise, khaki shorts and white trainers. The other people in this view are participant B, a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks and participant C, an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with short messy brown hair, wearing a hoodie in bright lime green, black joggers and white socks, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with a dark bob, wearing a long-sleeved T-shirt in vivid magenta pink, grey lounge trousers and black slippers, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f4fcc8397cfa870ec70c029846dda06d8afb8ca1
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots. The other participants of the same recording are participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes; participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 4> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4898d95ac2440526a955e8b3ac222b9a8b35736e
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots. The other participants of the same recording are participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes; participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 4> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/white/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4f1f48773d200c82bf04c5f324475e086e84a31e
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots. The other participants of the same recording are participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes; participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 4> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the upper part of the frame, a small figure, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e2b6f768a42d5aafbcc24034f6bf376caa68cae0
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant A in the foreground covers participant C from about 0.8 s to 1.9 s; participant A in the foreground covers participant D from about 0.0 s to 1.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d4c3872cf8c112fa4dbcbe2315d87f696397d8fd
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant A in the foreground covers participant C from about 0.8 s to 1.9 s; participant A in the foreground covers participant D from about 0.0 s to 1.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/white/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c71fe716e8a3855776397d386f85a0c3594f79fc
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the upper part of the frame, a small figure, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant A in the foreground covers participant C from about 0.8 s to 1.9 s; participant A in the foreground covers participant D from about 0.0 s to 1.0 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ab6a3b3568a29f649b1accc319d2b6162f67b6c9
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands in the centre of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2913cd8fdbc3c9ed76081124dc26729597a405d6
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands in the centre of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/white/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..23402f0a168ab7a6de78f9c3c902cfc330479e8e
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-c/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes and participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the upper part of the frame, tiny in the distance, in the far background, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands in the centre of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2e1e30bf15a0fda8671afdc4b0497f62d979cc20
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes and participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 5> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 5>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant C) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The storage rack with varied stock and empty spaces in the foreground covers participant B from about 3.6 s to 3.8 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..84d86b709fb8fc407f8d8bc7c8a71c99aebf9d3d
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes and participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 5> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 5>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant C) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The storage rack with varied stock and empty spaces in the foreground covers participant B from about 3.6 s to 3.8 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/white/prompt.txt b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8d063779b36823eca2a1a3754e96ae6310a4755c
--- /dev/null
+++ b/prompts/multi_view/multi4-020-warehouse-opposing-pairs/follow-d/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking along a warehouse aisle as two pairs of people pass each other in opposite directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a roofed warehouse: tall steel racks stocked with boxes and crates line both sides of a wide aisle, a few handcarts stand at the rack ends, a high ceiling with hanging lights spans the hall, and the aisle runs on between more racks. The people walk along the open aisle. Preserve the reference spacing, rack and crate positions and open walking routes. Use sealed grey concrete, painted steel racking, cardboard boxes and raw timber crates. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a slim adult with short dark curls, wearing a work shirt in bright turquoise, black cargo trousers and black safety boots. The other participants of the same recording are participant A, an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots; participant B, a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes and participant C, a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is an adult of medium build with short black hair, wearing a work fleece in royal blue, grey cargo trousers and black safety boots, and keeps exactly this appearance. <Subject 4> (participant B) is a tall adult with a light-brown ponytail, wearing a high-visibility vest in bright lime green over a white T-shirt, black trousers and grey safety shoes, and keeps exactly this appearance. <Subject 5> (participant C) is a stocky adult with a shaved head, wearing a padded gilet in vivid magenta pink over a grey hoodie, khaki trousers and brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 5>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant B) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 5> (participant C) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and B walk one way along one lane of the aisle, B ahead of A, while participants C and D walk the opposite way along a parallel lane, D ahead of C, so the two pairs close in on each other; B and D pass each other with lateral clearance near the end of this span; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, the pairs keep walking without stopping or turning and pass one another with lateral clearance - A and D and B and C at about the same moment, then A and C - before moving apart in their own directions; participant A moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The storage rack with varied stock and empty spaces in the foreground covers participant B from about 3.6 s to 3.8 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0840c1fdbc45e3ed1cb82bb63ae453825c4396bc
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes. The other participants of the same recording are participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers; participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the side, facing the screen-right; <Subject 5> (participant D) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward the middle of the concourse along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant D in the foreground covers the protagonist from about 3.9 s to 4.2 s; participant D in the foreground covers participant B from about 2.7 s to 3.6 s; the protagonist (participant A) in the foreground covers participant C from about 0.9 s to 2.3 s; participant D in the foreground covers participant C from about 3.3 s to 3.8 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e5fafc6da82b02cb579395d2f2e16886607d103b
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes. The other participants of the same recording are participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers; participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the side, facing the screen-right; <Subject 5> (participant D) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward the middle of the concourse along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant D in the foreground covers the protagonist from about 3.9 s to 4.2 s; participant D in the foreground covers participant B from about 2.7 s to 3.6 s; the protagonist (participant A) in the foreground covers participant C from about 0.9 s to 2.3 s; participant D in the foreground covers participant C from about 3.3 s to 3.8 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-a/white/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4436ac6e09b9a8d855be9fd842037359a0d40a39
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes. The other participants of the same recording are participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers; participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands on the screen-left of the upper part of the frame, a small figure, behind <Subject 5>, directly behind <Subject 1>, seen from the side, facing the screen-right; <Subject 5> (participant D) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward the middle of the concourse along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; the protagonist (participant A) moves about 3 metres along its route; participant B moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant D in the foreground covers the protagonist from about 3.9 s to 4.2 s; participant D in the foreground covers participant B from about 2.7 s to 3.6 s; the protagonist (participant A) in the foreground covers participant C from about 0.9 s to 2.3 s; participant D in the foreground covers participant C from about 3.3 s to 3.8 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f4fdd8cec75322f1b5456a68fe01d3f8f96af0e8
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers. The other participants of the same recording are participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes; participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from the side, facing the screen-right. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward the middle of the concourse along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant C in the foreground covers participant A from about 2.7 s to 3.4 s; participant C in the foreground covers the protagonist from about 3.8 s to 4.2 s; the protagonist (participant B) in the foreground covers participant D from about 0.9 s to 2.3 s; participant C in the foreground covers participant D from about 3.2 s to 3.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..725fd8a83262e600ff14e1aee25160e579f35cc4
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers. The other participants of the same recording are participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes; participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from the side, facing the screen-right. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward the middle of the concourse along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant C in the foreground covers participant A from about 2.7 s to 3.4 s; participant C in the foreground covers the protagonist from about 3.8 s to 4.2 s; the protagonist (participant B) in the foreground covers participant D from about 0.9 s to 2.3 s; participant C in the foreground covers participant D from about 3.2 s to 3.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-b/white/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..eaf6401e1bd4c017a7de61a990634c70143cb2ff
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse where the participants' routes cross from four directions, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers. The other participants of the same recording are participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes; participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. <Subject 5> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-right of the upper part of the frame, a small figure, in the far background, to the screen-right of <Subject 1>, seen from the front, facing toward the camera; <Subject 4> (participant C) stands at the far screen-right edge of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, partly cut off by the frame edge, seen from the side, facing the screen-left; <Subject 5> (participant D) stands in the centre of the upper part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from the side, facing the screen-right. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, the participants walk toward the middle of the concourse along offset straight routes from different directions - participant A heading north, B south, C east and D west - so their routes converge on the crossing; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, they cross through the middle with clearance, A passing closest to D and B passing closest to C, and each keeps walking straight on along their own direction, moving apart; participant A moves about 3 metres along its route; the protagonist (participant B) moves about 3 metres along its route; participant C moves about 3 metres along its route; participant D moves about 3 metres along its route. Participant C in the foreground covers participant A from about 2.7 s to 3.4 s; participant C in the foreground covers the protagonist from about 3.8 s to 4.2 s; the protagonist (participant B) in the foreground covers participant D from about 0.9 s to 2.3 s; participant C in the foreground covers participant D from about 3.2 s to 3.7 s. Keep the reference's recorded vertical field of view at every frame. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..00e7ca5b1433a3a44fb1dad9af104c8aa49190cf
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant B), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots. The other people in this view are participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers, and keeps exactly this appearance. <Subject 4> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. The protagonist (participant C) in the foreground covers participant B from about 1.3 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5bc72513b0b03509ef642601ef221b905143afc8
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-c/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant B), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots. The other people in this view are participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers, and keeps exactly this appearance. <Subject 4> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. The protagonist (participant C) in the foreground covers participant B from about 1.3 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-c/white/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ec4b3edc12fe8b7362bc9fe9dbc7d1671e590c41
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-c/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant B), <Subject 4> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots. The other people in this view are participant B, a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers and participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a young adult with a short blond crop, wearing a puffer jacket in bright lime green, black jeans and white sneakers, and keeps exactly this appearance. <Subject 4> (participant D) is a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant D) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 3 metres along its route; the protagonist (participant C) moves about 3 metres along its route; participant D moves about 3 metres along its route. The protagonist (participant C) in the foreground covers participant B from about 1.3 s to 2.3 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..bca92fda6141678cce75bf1dc7acc08fac511f75
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout where purple blocks are buildings (walls, roofs and railings included), orange-brown blocks are tables and cabinets, grey-blue strips are roads, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes. The other people in this view are participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes and participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant C) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The protagonist (participant D) in the foreground covers participant A from about 1.1 s to 2.2 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..128aae80c0d7e5b0ebb0ce58f271049332f83273
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-d/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes. The other people in this view are participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes and participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant C) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The protagonist (participant D) in the foreground covers participant A from about 1.1 s to 2.2 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-021-transit-crossroads/follow-d/white/prompt.txt b/prompts/multi_view/multi4-021-transit-crossroads/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..46713b1a2479de9239f46f7e354248f5adc96831
--- /dev/null
+++ b/prompts/multi_view/multi4-021-transit-crossroads/follow-d/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant A), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking across a transit station concourse, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: An open transit station concourse: a broad canopy roof on tall pillars shelters part of a wide, level paved floor, platforms, ticket gates and waiting benches stand along its edges, and the concourse opens toward the tracks and the sky. The people walk on the open floor. Preserve the reference spacing, pillar, platform and bench positions and open walking routes. Use polished stone paving, painted steel pillars, concrete platform edges and glass. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a stocky adult with a grey beard, wearing a zip-up fleece in bright turquoise, dark grey trousers and brown walking shoes. The other people in this view are participant A, a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes and participant C, an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a tall adult with short dark hair, wearing a wool overcoat in royal blue, grey trousers and black leather shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with long dark hair, wearing a raincoat in vivid magenta pink, beige trousers and black ankle boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, a small figure in the middle distance, behind <Subject 1>, to the screen-left of <Subject 1>, seen from the side, facing the screen-right; <Subject 4> (participant C) stands on the screen-right of the middle part of the frame, a small figure in the middle distance, behind <Subject 3>, to the screen-right of <Subject 1>, seen from the front, facing toward the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant A moves about 3 metres along its route; participant C moves about 3 metres along its route; the protagonist (participant D) moves about 3 metres along its route. The protagonist (participant D) in the foreground covers participant A from about 1.1 s to 2.2 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-a/colored-obb/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-a/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fd7263e7281ecc5b4ff98f3057f79a57f71a04e2
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-a/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout where grey-brown shapes are rocks, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots. The other participants of the same recording are participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes; participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-a/mixed/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-a/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..58f3a6cc3eb92780809710c608319ba7d93b3fcc
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-a/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots. The other participants of the same recording are participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes; participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-a/white/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-a/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e07677baf7f7f1724f4532e6bcbd3324697e13c5
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-a/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant A), <Subject 3> (participant B), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant A); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots. The other participants of the same recording are participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes; participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, nearest to the camera, seen from behind, facing away from the camera; <Subject 3> (participant B) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands in the centre of the middle part of the frame, a small figure, behind <Subject 3>, directly behind <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, a small figure, behind <Subject 4>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; the protagonist (participant A) moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-b/colored-obb/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-b/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ba9607ae0c8746c854e71114851e8f284d3483d2
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-b/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout where grey-brown shapes are rocks, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7 that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes. The other participants of the same recording are participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots; participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, nearest to the camera, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. The irregular faceted rock in the foreground covers participant A from about 2.2 s to 3.6 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-b/mixed/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-b/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e7bd6ab192e4fc620386cae1f0497497709b8d72
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-b/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes. The other participants of the same recording are participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots; participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, nearest to the camera, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. The irregular faceted rock in the foreground covers participant A from about 2.2 s to 3.6 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-b/white/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-b/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5e11a8d19db1e7a93d5f6c888de1e9b8d5216d79
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-b/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant B), <Subject 3> (participant A), <Subject 4> (participant C), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant B); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes. The other participants of the same recording are participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots; participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, nearest to the camera, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, a small figure, behind <Subject 1>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera; <Subject 5> (participant D) stands in the centre of the middle part of the frame, a small figure, behind <Subject 4>, directly behind <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; the protagonist (participant B) moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. The irregular faceted rock in the foreground covers participant A from about 2.2 s to 3.6 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-c/colored-obb/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-c/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2d1ea69c49b41fec33c80976c957b81eea26fd1f
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-c/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 4> is not in frame at 00:00.000 and enters at about 0.6 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the human proxy tagged #f736df; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout where grey-brown shapes are rocks, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the human proxy tagged #366ef7; use the target appearance described below.
+<Subject 4> is participant B: not in frame at 00:00.000; it enters the frame at about 0.6 s; in <Video 1> the human proxy tagged #a7f736 that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the human proxy tagged #36f7d6 that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (enters [Shot 1] at about 0.6 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots. The other participants of the same recording are participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots; participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots, and keeps exactly this appearance. <Subject 4> (participant B) is not in frame at 00:00.000 and enters the frame at about 0.6 s; from then on it is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-c/mixed/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-c/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1d81767c284a04bbfbb4ea05e2fec298168dfe7b
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-c/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 4> is not in frame at 00:00.000 and enters at about 0.6 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant B: not in frame at 00:00.000; it enters the frame at about 0.6 s; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the blue mannequin proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (enters [Shot 1] at about 0.6 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots. The other participants of the same recording are participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots; participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots, and keeps exactly this appearance. <Subject 4> (participant B) is not in frame at 00:00.000 and enters the frame at about 0.6 s; from then on it is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-c/white/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-c/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5ad9b2f24ca5a04484fbcd1adb95b9be4b57e591
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-c/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist, participant C), <Subject 3> (participant A), <Subject 4> (participant B), <Subject 5> (participant D); nothing else in <Video 1> is a person or an interaction object. <Subject 4> is not in frame at 00:00.000 and enters at about 0.6 s.
+<Subject 1> is the protagonist (participant C); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant A; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant B: not in frame at 00:00.000; it enters the frame at about 0.6 s; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+<Subject 5> is participant D; in <Video 1> the grey block-figure proxy that starts on the right of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave as two faster people overtake two slower ones, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (enters [Shot 1] at about 0.6 s): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target appearance described below and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots. The other participants of the same recording are participant A, a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots; participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes and participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant A) is a stocky adult with short dark hair, wearing a caving jacket in royal blue, grey trousers, a white helmet and black boots, and keeps exactly this appearance. <Subject 4> (participant B) is not in frame at 00:00.000 and enters the frame at about 0.6 s; from then on it is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 5> (participant D) is a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, seen from behind, facing away from the camera; <Subject 3> (participant A) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 5> (participant D) stands on the screen-right of the middle part of the frame, about a third to half of the frame height, behind <Subject 1>, to the screen-right of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, participants A and C walk uphill forward slowly one behind the other along the cave floor while participants B and D start level with them on a parallel lane - B beside A, D beside C - and walk uphill much faster, pulling ahead of A and C; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, b moves past C, and both faster participants keep pulling further ahead while A and C continue at their slow, steady pace; participant A moves about 2 metres along its route; participant B moves about 4 metres along its route, rises about half a metre; the protagonist (participant C) moves about 2 metres along its route; participant D moves about 4 metres along its route, rises a short distance. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-d/colored-obb/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-d/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ae910ea90d90c1d6065e1613cf4d50ea103c368e
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-d/colored-obb/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the human proxy tagged #36f7d6; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout where grey-brown shapes are rocks, the sky is blue and the ground is dark grey; these solid colours are semantic class codes shared by every entity of a class, not materials; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the human proxy tagged #a7f736; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the human proxy tagged #f736df that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots. The other people in this view are participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes and participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route, rises a short distance. The irregular faceted rock in the foreground covers participant C from about 3.6 s to 3.9 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-d/mixed/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-d/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4f19d37606f3fbaea69714e8d430ef138f780ceb
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-d/mixed/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as green low-poly trees and black blocks on a dark grey ground under a blue sky; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the blue mannequin proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the blue mannequin proxy that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and its articulated motion from the timeline below.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots. The other people in this view are participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes and participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route, rises a short distance. The irregular faceted rock in the foreground covers participant C from about 3.6 s to 3.9 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/multi_view/multi4-023-cave-overtake/follow-d/white/prompt.txt b/prompts/multi_view/multi4-023-cave-overtake/follow-d/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ab35b4c490d6ff0d6c1011466ff9c627a04180e1
--- /dev/null
+++ b/prompts/multi_view/multi4-023-cave-overtake/follow-d/white/prompt.txt
@@ -0,0 +1,36 @@
+AWM_PROXY_CONTROL. One visual reference accompanies this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. Each subject's position and motion come from <Video 1>. Every subject's appearance comes from the text below, rendered consistently in the worldview and art direction it describes. The whole take is rendered in this style, from the first frame to the last: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. There is no box, frame, glass panel, outline or coloured slab around or in front of any person, animal or object in the output; every figure stands directly in the scene.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style, and every subject's identity, appearance and materials, follow the text; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. The first frame of the target reproduces frame 0 of <Video 1> in the appearance the text describes, and the clip continues directly from it while performing the timestamped actions.
+
+subject_definitions:
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist, participant D), <Subject 3> (participant B), <Subject 4> (participant C); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist (participant D); in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 2> is the environment and its fixed spatial arrangement: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. In <Video 1> the same layout as grey low-poly trees and blocks on a dark grey ground; these shapes and colours are approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is participant B; in <Video 1> the grey block-figure proxy; use the target appearance described below.
+<Subject 4> is participant C; in <Video 1> the grey block-figure proxy that starts on the left of the frame; use the target appearance described below.
+
+summary:
+[reference generation] Generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> walking up the sloping floor of a natural cave, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and realize the scene with the detailed target materials below.
+
+retention_analysis:
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from the target materials below.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from the target materials below and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] <Subject 2>, the setting: The inside of a large natural cave: irregular rock walls and thick rock pillars rise on both sides, a low, uneven rock ceiling hangs overhead with an opening to daylight near the entrance, and the rocky floor rises steadily ahead. The people walk up the open cave floor. Preserve the reference spacing, wall and pillar positions, the rising ground and open walking routes. Use rough grey limestone, damp dark rock, loose gravel and dim natural light. <Subject 1>, the protagonist, has the following target appearance: The protagonist is participant D, a wiry adult with a shaved head, wearing a softshell jacket in bright turquoise, charcoal trousers, a grey helmet and black boots. The other people in this view are participant B, a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes and participant C, an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots; only those listed as subjects above appear in this view. Each person keeps one consistent face, build and outfit for the whole clip, no two people share the same outfit and nobody carries anything. <Subject 3> (participant B) is a tall lean adult with a blond ponytail, wearing a long-sleeved base layer in bright lime green, black trousers, a black helmet and grey approach shoes, and keeps exactly this appearance. <Subject 4> (participant C) is an adult of average build with short curly brown hair, wearing a fleece in vivid magenta pink, khaki trousers, a white helmet and brown boots, and keeps exactly this appearance. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. At the start of the clip <Subject 1> (the protagonist) stands in the centre of the middle part of the frame, about a third to half of the frame height, behind <Subject 4>, seen from behind, facing away from the camera; <Subject 3> (participant B) stands in the centre of the lower part of the frame, about a third to half of the frame height, nearest to the camera, partly cut off by the frame edge, seen from behind, facing away from the camera; <Subject 4> (participant C) stands on the screen-left of the middle part of the frame, about a third to half of the frame height, behind <Subject 3>, to the screen-left of <Subject 1>, seen from behind, facing away from the camera. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.500, everyone shown moves steadily along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route, rises about half a metre. From 00:02.500 to 00:05.000, everyone shown keeps moving along their own route exactly as recorded in <Video 1>; participant B moves about 4 metres along its route, rises about half a metre; participant C moves about 2 metres along its route; the protagonist (participant D) moves about 4 metres along its route, rises a short distance. The irregular faceted rock in the foreground covers participant C from about 3.6 s to 3.9 s. Keep the reference's fixed 60-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. Rendering style, stated again because it governs every frame: Realistic live-action footage of the place described below in natural light suited to it, with natural human anatomy and motion, believable scale, detailed physical materials and consistent shadows; no proxy colours, wireframes, block figures or game UI. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+N/A
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7c6018d212f3f5f43156d11f68838b78a050c772
--- /dev/null
+++ b/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a tense standoff on a concrete airport apron where a protagonist in tactical gear faces two armed men near large hangars, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with short hair, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers and tan boots, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings, large grey industrial hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a sage bomber jacket, a white t-shirt, blue jeans and white sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a grey hoodie, black track pants with white stripes and grey sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings, large grey industrial hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short hair, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers and tan boots, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a sage bomber jacket, a white t-shirt, blue jeans and white sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a grey hoodie, black track pants with white stripes and grey sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 50.124-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/mixed/prompt.txt b/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3d141bdc90bbf9b6012b68d9eb9717d430824590
--- /dev/null
+++ b/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a tense standoff on a concrete airport apron where a protagonist in tactical gear faces two armed men near large hangars, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short hair, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers and tan boots, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings, large grey industrial hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a sage bomber jacket, a white t-shirt, blue jeans and white sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a grey hoodie, black track pants with white stripes and grey sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings, large grey industrial hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short hair, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers and tan boots, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a sage bomber jacket, a white t-shirt, blue jeans and white sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a grey hoodie, black track pants with white stripes and grey sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 50.124-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/white/prompt.txt b/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..692a1a1c4edce0f18d826b882d0d79b1295a597d
--- /dev/null
+++ b/prompts/neon/ndm-01-cam-dynamic_obj-static-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a tense standoff on a concrete airport apron where a protagonist in tactical gear faces two armed men near large hangars, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short hair, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers and tan boots, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings, large grey industrial hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a sage bomber jacket, a white t-shirt, blue jeans and white sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a grey hoodie, black track pants with white stripes and grey sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings, large grey industrial hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short hair, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers and tan boots, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a sage bomber jacket, a white t-shirt, blue jeans and white sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a grey hoodie, black track pants with white stripes and grey sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 50.124-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..975b283f357a719629cdfbeb467a3eda98cd4255
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald man in a tactical vest stands on an airport tarmac facing two armed men who are aiming weapons at him, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a bald adult male seen from behind, wearing a grey quilted tactical vest over a black long-sleeved shirt, camouflage trousers, and black boots. He stands with a relaxed posture, facing away from the camera; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide concrete airport apron marked with yellow taxiway lines and moss-toned painted safety zones. To the left is a large industrial hangar with segmented metal doors. In the background, there is a tall control tower with a glazed observation deck, stacks of shipping containers in rust and navy, and parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a brown leather jacket, a white shirt, khaki trousers, and dark shoes. He stands facing the camera, holding a handgun with both hands at chest level; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a navy blazer, a white shirt, dark trousers, and black shoes. He stands facing the camera, holding a handgun with both hands at chest level; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide concrete airport apron marked with yellow taxiway lines and moss-toned painted safety zones. To the left is a large industrial hangar with segmented metal doors. In the background, there is a tall control tower with a glazed observation deck, stacks of shipping containers in rust and navy, and parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald adult male seen from behind, wearing a grey quilted tactical vest over a black long-sleeved shirt, camouflage trousers, and black boots. He stands with a relaxed posture, facing away from the camera, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a brown leather jacket, a white shirt, khaki trousers, and dark shoes. He stands facing the camera, holding a handgun with both hands at chest level, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a navy blazer, a white shirt, dark trousers, and black shoes. He stands facing the camera, holding a handgun with both hands at chest level, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/mixed/prompt.txt b/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..13699923b4b51aec11c3752107b91b24ce4eaf92
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald man in a tactical vest stands on an airport tarmac facing two armed men who are aiming weapons at him, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a bald adult male seen from behind, wearing a grey quilted tactical vest over a black long-sleeved shirt, camouflage trousers, and black boots. He stands with a relaxed posture, facing away from the camera; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide concrete airport apron marked with yellow taxiway lines and moss-toned painted safety zones. To the left is a large industrial hangar with segmented metal doors. In the background, there is a tall control tower with a glazed observation deck, stacks of shipping containers in rust and navy, and parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a brown leather jacket, a white shirt, khaki trousers, and dark shoes. He stands facing the camera, holding a handgun with both hands at chest level; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a navy blazer, a white shirt, dark trousers, and black shoes. He stands facing the camera, holding a handgun with both hands at chest level; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide concrete airport apron marked with yellow taxiway lines and moss-toned painted safety zones. To the left is a large industrial hangar with segmented metal doors. In the background, there is a tall control tower with a glazed observation deck, stacks of shipping containers in rust and navy, and parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald adult male seen from behind, wearing a grey quilted tactical vest over a black long-sleeved shirt, camouflage trousers, and black boots. He stands with a relaxed posture, facing away from the camera, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a brown leather jacket, a white shirt, khaki trousers, and dark shoes. He stands facing the camera, holding a handgun with both hands at chest level, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a navy blazer, a white shirt, dark trousers, and black shoes. He stands facing the camera, holding a handgun with both hands at chest level, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/white/prompt.txt b/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..263b661e10c29504ed8e0d7642860594ee1f5701
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-dynamic_obj-static-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald man in a tactical vest stands on an airport tarmac facing two armed men who are aiming weapons at him, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a bald adult male seen from behind, wearing a grey quilted tactical vest over a black long-sleeved shirt, camouflage trousers, and black boots. He stands with a relaxed posture, facing away from the camera; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide concrete airport apron marked with yellow taxiway lines and moss-toned painted safety zones. To the left is a large industrial hangar with segmented metal doors. In the background, there is a tall control tower with a glazed observation deck, stacks of shipping containers in rust and navy, and parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a brown leather jacket, a white shirt, khaki trousers, and dark shoes. He stands facing the camera, holding a handgun with both hands at chest level; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a navy blazer, a white shirt, dark trousers, and black shoes. He stands facing the camera, holding a handgun with both hands at chest level; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide concrete airport apron marked with yellow taxiway lines and moss-toned painted safety zones. To the left is a large industrial hangar with segmented metal doors. In the background, there is a tall control tower with a glazed observation deck, stacks of shipping containers in rust and navy, and parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald adult male seen from behind, wearing a grey quilted tactical vest over a black long-sleeved shirt, camouflage trousers, and black boots. He stands with a relaxed posture, facing away from the camera, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a brown leather jacket, a white shirt, khaki trousers, and dark shoes. He stands facing the camera, holding a handgun with both hands at chest level, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a navy blazer, a white shirt, dark trousers, and black shoes. He stands facing the camera, holding a handgun with both hands at chest level, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..89db95bc8d6bf49a7084f262dabcc8324d505bd1
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff on a concrete airport apron, with a control tower and hangars in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair and a beard, wearing an olive jacket, black cargo trousers, and brown boots, standing with his back to the camera and aiming a handgun forward; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and embedded blue lights, surrounded by large industrial hangars with segmented doors, a tall control tower with a glazed observation cabin, stacked cargo containers, and floodlight masts, set against distant mountain ridges under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a black sleeveless shirt, dark trousers, and black shoes, standing in a defensive stance and aiming a handgun; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a maroon plaid shirt, blue jeans, and brown boots, standing in a defensive stance and aiming a handgun; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and embedded blue lights, surrounded by large industrial hangars with segmented doors, a tall control tower with a glazed observation cabin, stacked cargo containers, and floodlight masts, set against distant mountain ridges under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair and a beard, wearing an olive jacket, black cargo trousers, and brown boots, standing with his back to the camera and aiming a handgun forward, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a black sleeveless shirt, dark trousers, and black shoes, standing in a defensive stance and aiming a handgun, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a maroon plaid shirt, blue jeans, and brown boots, standing in a defensive stance and aiming a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves a short distance along its route, lowers the hands to the sides. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 2 metres along its route, lowers the hands to the sides. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c7038f173dddbcc14020ce5c0b03c84241b6cb03
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff on a concrete airport apron, with a control tower and hangars in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair and a beard, wearing an olive jacket, black cargo trousers, and brown boots, standing with his back to the camera and aiming a handgun forward; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and embedded blue lights, surrounded by large industrial hangars with segmented doors, a tall control tower with a glazed observation cabin, stacked cargo containers, and floodlight masts, set against distant mountain ridges under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a black sleeveless shirt, dark trousers, and black shoes, standing in a defensive stance and aiming a handgun; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a maroon plaid shirt, blue jeans, and brown boots, standing in a defensive stance and aiming a handgun; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and embedded blue lights, surrounded by large industrial hangars with segmented doors, a tall control tower with a glazed observation cabin, stacked cargo containers, and floodlight masts, set against distant mountain ridges under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair and a beard, wearing an olive jacket, black cargo trousers, and brown boots, standing with his back to the camera and aiming a handgun forward, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a black sleeveless shirt, dark trousers, and black shoes, standing in a defensive stance and aiming a handgun, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a maroon plaid shirt, blue jeans, and brown boots, standing in a defensive stance and aiming a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves a short distance along its route, lowers the hands to the sides. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 2 metres along its route, lowers the hands to the sides. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..86fbbad7ede1fe2fdb9e4df377d680b443f8513a
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-static_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff on a concrete airport apron, with a control tower and hangars in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair and a beard, wearing an olive jacket, black cargo trousers, and brown boots, standing with his back to the camera and aiming a handgun forward; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and embedded blue lights, surrounded by large industrial hangars with segmented doors, a tall control tower with a glazed observation cabin, stacked cargo containers, and floodlight masts, set against distant mountain ridges under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a black sleeveless shirt, dark trousers, and black shoes, standing in a defensive stance and aiming a handgun; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair, wearing a maroon plaid shirt, blue jeans, and brown boots, standing in a defensive stance and aiming a handgun; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and embedded blue lights, surrounded by large industrial hangars with segmented doors, a tall control tower with a glazed observation cabin, stacked cargo containers, and floodlight masts, set against distant mountain ridges under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair and a beard, wearing an olive jacket, black cargo trousers, and brown boots, standing with his back to the camera and aiming a handgun forward, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a black sleeveless shirt, dark trousers, and black shoes, standing in a defensive stance and aiming a handgun, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair, wearing a maroon plaid shirt, blue jeans, and brown boots, standing in a defensive stance and aiming a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves a short distance along its route, lowers the hands to the sides. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 2 metres along its route, lowers the hands to the sides. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1878eb90da4dc0135972537a1518d3a182dbb40a
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand on a concrete airport apron facing each other in a standoff, with a large hangar, a control tower, and a parked passenger jet in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 1.9 s.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket, black trousers, and brown boots, standing with his back to the camera and holding a handgun with both hands; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by a large industrial hangar with segmented doors on the left and a concrete control tower with a glazed observation cabin on the right. A white passenger aircraft is parked near the hangar, and stacked cargo containers sit near the tower. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair, wearing a grey hoodie, black track pants with white stripes, and sneakers, standing in the middle distance facing the camera and holding a handgun; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a brown leather jacket, tan trousers, and dark shoes, standing in the middle distance facing the camera and holding a handgun; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+<Subject 5> is armed person 3: appearance is not established by <Picture 1>; it enters the frame at about 1.9 s; in <Video 1> the human proxy tagged #cdf47b that starts on the left of the frame, whose white face is its front; preserve its recorded identity without inferring unseen clothing or materials from the image.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 3 down in an armed standoff as a newcomer walks in and the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (enters [Shot 1] at about 1.9 s): partially_preserved - retain identity, geometry, trajectory and interaction timing from <Video 1>; appearance is not established by <Picture 1>, so do not claim to preserve unseen details from that image.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by a large industrial hangar with segmented doors on the left and a concrete control tower with a glazed observation cabin on the right. A white passenger aircraft is parked near the hangar, and stacked cargo containers sit near the tower. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket, black trousers, and brown boots, standing with his back to the camera and holding a handgun with both hands, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair, wearing a grey hoodie, black track pants with white stripes, and sneakers, standing in the middle distance facing the camera and holding a handgun, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a brown leather jacket, tan trousers, and dark shoes, standing in the middle distance facing the camera and holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, has no discernible appearance in <Picture 1>; preserve its recorded identity and motion from <Video 1> without treating unspecified appearance as image evidence. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:03.267, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they fire back while another armed person walks into view; armed person 3 moves about 5 metres along its route. From 00:03.267 to 00:05.000, the protagonist shoots armed person 3, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 3 drops about half a metre. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind blowing across the tarmac, punctuated by the sharp metallic click of a pistol slide being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/mixed/prompt.txt b/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..442fa9302aca4176b24f9adce8a86331818736f0
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/mixed/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand on a concrete airport apron facing each other in a standoff, with a large hangar, a control tower, and a parked passenger jet in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 1.9 s.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket, black trousers, and brown boots, standing with his back to the camera and holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by a large industrial hangar with segmented doors on the left and a concrete control tower with a glazed observation cabin on the right. A white passenger aircraft is parked near the hangar, and stacked cargo containers sit near the tower. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair, wearing a grey hoodie, black track pants with white stripes, and sneakers, standing in the middle distance facing the camera and holding a handgun; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a brown leather jacket, tan trousers, and dark shoes, standing in the middle distance facing the camera and holding a handgun; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: appearance is not established by <Picture 1>; it enters the frame at about 1.9 s; in <Video 1> the blue mannequin proxy that starts on the left of the frame; preserve its recorded identity without inferring unseen clothing or materials from the image.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 3 down in an armed standoff as a newcomer walks in and the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (enters [Shot 1] at about 1.9 s): partially_preserved - retain identity, geometry, trajectory and interaction timing from <Video 1>; appearance is not established by <Picture 1>, so do not claim to preserve unseen details from that image.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by a large industrial hangar with segmented doors on the left and a concrete control tower with a glazed observation cabin on the right. A white passenger aircraft is parked near the hangar, and stacked cargo containers sit near the tower. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket, black trousers, and brown boots, standing with his back to the camera and holding a handgun with both hands, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair, wearing a grey hoodie, black track pants with white stripes, and sneakers, standing in the middle distance facing the camera and holding a handgun, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a brown leather jacket, tan trousers, and dark shoes, standing in the middle distance facing the camera and holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, has no discernible appearance in <Picture 1>; preserve its recorded identity and motion from <Video 1> without treating unspecified appearance as image evidence. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:03.267, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they fire back while another armed person walks into view; armed person 3 moves about 5 metres along its route. From 00:03.267 to 00:05.000, the protagonist shoots armed person 3, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 3 drops about half a metre. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind blowing across the tarmac, punctuated by the sharp metallic click of a pistol slide being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/white/prompt.txt b/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..cec410f13b0cefb469a33476c0e3020bab691487
--- /dev/null
+++ b/prompts/neon/ndm-02-cam-static_obj-enter-death/follow/white/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand on a concrete airport apron facing each other in a standoff, with a large hangar, a control tower, and a parked passenger jet in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object. <Subject 5> is not in frame at 00:00.000 and enters at about 1.9 s.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket, black trousers, and brown boots, standing with his back to the camera and holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by a large industrial hangar with segmented doors on the left and a concrete control tower with a glazed observation cabin on the right. A white passenger aircraft is parked near the hangar, and stacked cargo containers sit near the tower. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair, wearing a grey hoodie, black track pants with white stripes, and sneakers, standing in the middle distance facing the camera and holding a handgun; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a brown leather jacket, tan trousers, and dark shoes, standing in the middle distance facing the camera and holding a handgun; in <Video 1> the grey block-figure proxy.
+<Subject 5> is armed person 3: appearance is not established by <Picture 1>; it enters the frame at about 1.9 s; in <Video 1> the grey block-figure proxy that starts on the left of the frame; preserve its recorded identity without inferring unseen clothing or materials from the image.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 3 down in an armed standoff as a newcomer walks in and the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (enters [Shot 1] at about 1.9 s): partially_preserved - retain identity, geometry, trajectory and interaction timing from <Video 1>; appearance is not established by <Picture 1>, so do not claim to preserve unseen details from that image.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by a large industrial hangar with segmented doors on the left and a concrete control tower with a glazed observation cabin on the right. A white passenger aircraft is parked near the hangar, and stacked cargo containers sit near the tower. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket, black trousers, and brown boots, standing with his back to the camera and holding a handgun with both hands, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair, wearing a grey hoodie, black track pants with white stripes, and sneakers, standing in the middle distance facing the camera and holding a handgun, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a brown leather jacket, tan trousers, and dark shoes, standing in the middle distance facing the camera and holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, has no discernible appearance in <Picture 1>; preserve its recorded identity and motion from <Video 1> without treating unspecified appearance as image evidence. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:03.267, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they fire back while another armed person walks into view; armed person 3 moves about 5 metres along its route. From 00:03.267 to 00:05.000, the protagonist shoots armed person 3, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 3 drops about half a metre. Keep the reference's fixed 42.6-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind blowing across the tarmac, punctuated by the sharp metallic click of a pistol slide being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/colored-obb/prompt.txt b/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4c20f08188dae9d25789221dc8d65533ed6b91c7
--- /dev/null
+++ b/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/colored-obb/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands on an airport tarmac facing three armed adversaries who are aiming weapons at him, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest, olive drab trousers, and tan boots, holding a firearm in a ready stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide concrete airport apron marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked cargo containers in rust and navy hues, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a navy long-sleeved shirt and dark trousers, aiming a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man wearing a crimson plaid shirt and grey jeans, aiming a handgun with both hands; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+<Subject 5> is armed person 3: in <Picture 1> a man with a shaved head and beard, wearing a black sleeveless shirt and dark trousers, aiming a handgun with both hands; in <Video 1> the human proxy tagged #cdf47b, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide concrete airport apron marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked cargo containers in rust and navy hues, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest, olive drab trousers, and tan boots, holding a firearm in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a navy long-sleeved shirt and dark trousers, aiming a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man wearing a crimson plaid shirt and grey jeans, aiming a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with a shaved head and beard, wearing a black sleeveless shirt and dark trousers, aiming a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about a metre along its route. Keep the reference's fixed 51.569-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/mixed/prompt.txt b/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..85e6521650b9055092f684af860f97e0d0e3e484
--- /dev/null
+++ b/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/mixed/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands on an airport tarmac facing three armed adversaries who are aiming weapons at him, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest, olive drab trousers, and tan boots, holding a firearm in a ready stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide concrete airport apron marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked cargo containers in rust and navy hues, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a navy long-sleeved shirt and dark trousers, aiming a handgun with both hands; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man wearing a crimson plaid shirt and grey jeans, aiming a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with a shaved head and beard, wearing a black sleeveless shirt and dark trousers, aiming a handgun with both hands; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide concrete airport apron marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked cargo containers in rust and navy hues, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest, olive drab trousers, and tan boots, holding a firearm in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a navy long-sleeved shirt and dark trousers, aiming a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man wearing a crimson plaid shirt and grey jeans, aiming a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with a shaved head and beard, wearing a black sleeveless shirt and dark trousers, aiming a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about a metre along its route. Keep the reference's fixed 51.569-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/white/prompt.txt b/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c4b04eed01b2ad2809b5dd008f390b331c728911
--- /dev/null
+++ b/prompts/neon/ndm-03-cam-dynamic_obj-dynamic-death-composite-forward-turn-v1/follow/white/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands on an airport tarmac facing three armed adversaries who are aiming weapons at him, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest, olive drab trousers, and tan boots, holding a firearm in a ready stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide concrete airport apron marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked cargo containers in rust and navy hues, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a navy long-sleeved shirt and dark trousers, aiming a handgun with both hands; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man wearing a crimson plaid shirt and grey jeans, aiming a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with a shaved head and beard, wearing a black sleeveless shirt and dark trousers, aiming a handgun with both hands; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide concrete airport apron marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked cargo containers in rust and navy hues, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest, olive drab trousers, and tan boots, holding a firearm in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a navy long-sleeved shirt and dark trousers, aiming a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man wearing a crimson plaid shirt and grey jeans, aiming a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with a shaved head and beard, wearing a black sleeveless shirt and dark trousers, aiming a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about a metre along its route. Keep the reference's fixed 51.569-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ae272364ac475631ab1776626232697ab7c14680
--- /dev/null
+++ b/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands on a concrete airport apron facing three armed men who are aiming weapons at him, with large industrial hangars and mountains in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers, and tan work boots. He stands with a relaxed posture, facing the group of men; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features large, grey corrugated metal hangars with segmented roll-up doors. To the left, there are stacked cargo containers in rust-red and white. A tall floodlight mast stands centrally between the buildings. In the far distance, hazy mountain ridges rise against a bright, overcast sky. A passenger aircraft is partially visible on the far right; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing an olive bomber jacket over a grey t-shirt, dark denim jeans, and white sneakers. He holds a handgun with both hands, aiming it forward; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with short hair, wearing a brown leather jacket, dark trousers, and dark shoes. He stands slightly behind the others, holding a handgun with both hands; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair and a beard, wearing a grey hoodie, black track pants with a white stripe down the side, and grey sneakers. He holds a handgun with both hands, aiming it forward; in <Video 1> the human proxy tagged #cdf47b, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features large, grey corrugated metal hangars with segmented roll-up doors. To the left, there are stacked cargo containers in rust-red and white. A tall floodlight mast stands centrally between the buildings. In the far distance, hazy mountain ridges rise against a bright, overcast sky. A passenger aircraft is partially visible on the far right, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers, and tan work boots. He stands with a relaxed posture, facing the group of men, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing an olive bomber jacket over a grey t-shirt, dark denim jeans, and white sneakers. He holds a handgun with both hands, aiming it forward, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short hair, wearing a brown leather jacket, dark trousers, and dark shoes. He stands slightly behind the others, holding a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair and a beard, wearing a grey hoodie, black track pants with a white stripe down the side, and grey sneakers. He holds a handgun with both hands, aiming it forward, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about a metre along its route. Keep the reference's fixed 51.569-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind across the tarmac, punctuated by the sharp metallic click of weapon slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2dab6fe930c3d739e9334faf194334a691cde5e7
--- /dev/null
+++ b/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands on a concrete airport apron facing three armed men who are aiming weapons at him, with large industrial hangars and mountains in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers, and tan work boots. He stands with a relaxed posture, facing the group of men; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features large, grey corrugated metal hangars with segmented roll-up doors. To the left, there are stacked cargo containers in rust-red and white. A tall floodlight mast stands centrally between the buildings. In the far distance, hazy mountain ridges rise against a bright, overcast sky. A passenger aircraft is partially visible on the far right; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing an olive bomber jacket over a grey t-shirt, dark denim jeans, and white sneakers. He holds a handgun with both hands, aiming it forward; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short hair, wearing a brown leather jacket, dark trousers, and dark shoes. He stands slightly behind the others, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair and a beard, wearing a grey hoodie, black track pants with a white stripe down the side, and grey sneakers. He holds a handgun with both hands, aiming it forward; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features large, grey corrugated metal hangars with segmented roll-up doors. To the left, there are stacked cargo containers in rust-red and white. A tall floodlight mast stands centrally between the buildings. In the far distance, hazy mountain ridges rise against a bright, overcast sky. A passenger aircraft is partially visible on the far right, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers, and tan work boots. He stands with a relaxed posture, facing the group of men, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing an olive bomber jacket over a grey t-shirt, dark denim jeans, and white sneakers. He holds a handgun with both hands, aiming it forward, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short hair, wearing a brown leather jacket, dark trousers, and dark shoes. He stands slightly behind the others, holding a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair and a beard, wearing a grey hoodie, black track pants with a white stripe down the side, and grey sneakers. He holds a handgun with both hands, aiming it forward, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about a metre along its route. Keep the reference's fixed 51.569-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind across the tarmac, punctuated by the sharp metallic click of weapon slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6f8b783cd9b09f307508fb6422634b4017a1d881
--- /dev/null
+++ b/prompts/neon/ndm-03-cam-static_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands on a concrete airport apron facing three armed men who are aiming weapons at him, with large industrial hangars and mountains in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers, and tan work boots. He stands with a relaxed posture, facing the group of men; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features large, grey corrugated metal hangars with segmented roll-up doors. To the left, there are stacked cargo containers in rust-red and white. A tall floodlight mast stands centrally between the buildings. In the far distance, hazy mountain ridges rise against a bright, overcast sky. A passenger aircraft is partially visible on the far right; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing an olive bomber jacket over a grey t-shirt, dark denim jeans, and white sneakers. He holds a handgun with both hands, aiming it forward; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short hair, wearing a brown leather jacket, dark trousers, and dark shoes. He stands slightly behind the others, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair and a beard, wearing a grey hoodie, black track pants with a white stripe down the side, and grey sneakers. He holds a handgun with both hands, aiming it forward; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features large, grey corrugated metal hangars with segmented roll-up doors. To the left, there are stacked cargo containers in rust-red and white. A tall floodlight mast stands centrally between the buildings. In the far distance, hazy mountain ridges rise against a bright, overcast sky. A passenger aircraft is partially visible on the far right, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a black long-sleeved shirt, olive cargo trousers, and tan work boots. He stands with a relaxed posture, facing the group of men, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing an olive bomber jacket over a grey t-shirt, dark denim jeans, and white sneakers. He holds a handgun with both hands, aiming it forward, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short hair, wearing a brown leather jacket, dark trousers, and dark shoes. He stands slightly behind the others, holding a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair and a beard, wearing a grey hoodie, black track pants with a white stripe down the side, and grey sneakers. He holds a handgun with both hands, aiming it forward, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about a metre along its route. Keep the reference's fixed 51.569-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant airport machinery and wind across the tarmac, punctuated by the sharp metallic click of weapon slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..48aebb29f7db617fd7d7af9dd8f6ea71458e091c
--- /dev/null
+++ b/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist stands on an airport tarmac facing three armed men in a standoff, with hangars and aircraft in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket and black trousers, standing with his back to the camera facing the group; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and moss-toned painted zones, surrounded by large concrete hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers, and parked passenger aircraft under a sunset sky with distant mountains; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a crimson plaid shirt and blue jeans, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b that starts on the right of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a black sleeveless vest and dark trousers, holding a handgun; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair, wearing a moss-toned jacket over a white shirt and blue jeans, holding a handgun; in <Video 1> the human proxy tagged #cdf47b, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and moss-toned painted zones, surrounded by large concrete hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers, and parked passenger aircraft under a sunset sky with distant mountains, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket and black trousers, standing with his back to the camera facing the group, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a crimson plaid shirt and blue jeans, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a black sleeveless vest and dark trousers, holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair, wearing a moss-toned jacket over a white shirt and blue jeans, holding a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 39.634-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the metallic click of a weapon being drawn.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/mixed/prompt.txt b/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b006b2bce59c3bc1103850eb9cc5791d906c0226
--- /dev/null
+++ b/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/mixed/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist stands on an airport tarmac facing three armed men in a standoff, with hangars and aircraft in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket and black trousers, standing with his back to the camera facing the group; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and moss-toned painted zones, surrounded by large concrete hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers, and parked passenger aircraft under a sunset sky with distant mountains; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a crimson plaid shirt and blue jeans, holding a handgun with both hands; in <Video 1> the blue mannequin proxy that starts on the right of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a black sleeveless vest and dark trousers, holding a handgun; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair, wearing a moss-toned jacket over a white shirt and blue jeans, holding a handgun; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and moss-toned painted zones, surrounded by large concrete hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers, and parked passenger aircraft under a sunset sky with distant mountains, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket and black trousers, standing with his back to the camera facing the group, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a crimson plaid shirt and blue jeans, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a black sleeveless vest and dark trousers, holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair, wearing a moss-toned jacket over a white shirt and blue jeans, holding a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 39.634-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the metallic click of a weapon being drawn.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/white/prompt.txt b/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..64f8014f9f82babbdfe4ddac97c86c4231c79b20
--- /dev/null
+++ b/prompts/neon/ndm-04-cam-dynamic_obj-static-death/follow/white/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist stands on an airport tarmac facing three armed men in a standoff, with hangars and aircraft in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket and black trousers, standing with his back to the camera facing the group; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and moss-toned painted zones, surrounded by large concrete hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers, and parked passenger aircraft under a sunset sky with distant mountains; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a crimson plaid shirt and blue jeans, holding a handgun with both hands; in <Video 1> the grey block-figure proxy that starts on the right of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a black sleeveless vest and dark trousers, holding a handgun; in <Video 1> the grey block-figure proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair, wearing a moss-toned jacket over a white shirt and blue jeans, holding a handgun; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and moss-toned painted zones, surrounded by large concrete hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers, and parked passenger aircraft under a sunset sky with distant mountains, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket and black trousers, standing with his back to the camera facing the group, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a crimson plaid shirt and blue jeans, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a black sleeveless vest and dark trousers, holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair, wearing a moss-toned jacket over a white shirt and blue jeans, holding a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 39.634-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the metallic click of a weapon being drawn.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4cdb741137881843e3d84e6090485dc22f8bc5b6
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in an olive jacket stands in a paved industrial yard facing three armed adversaries, with warehouses and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a person seen from behind wearing an olive jacket, black trousers, and brown boots, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked concrete tarmac surface marked with mustard painted lines. The background features large corrugated metal warehouses with roll-up doors, a tall concrete observation tower, and stacked shipping containers in rust and navy hues. Distant mountains are visible under a bright, overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair wearing a brown leather jacket, khaki trousers, and black boots, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a grey hoodie, black track pants with white stripes, and grey sneakers, holding a handgun with both hands; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair wearing a navy jacket, black trousers, and black sneakers, holding a handgun with both hands; in <Video 1> the human proxy tagged #cdf47b that starts on the left of the frame, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked concrete tarmac surface marked with mustard painted lines. The background features large corrugated metal warehouses with roll-up doors, a tall concrete observation tower, and stacked shipping containers in rust and navy hues. Distant mountains are visible under a bright, overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a person seen from behind wearing an olive jacket, black trousers, and brown boots, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair wearing a brown leather jacket, khaki trousers, and black boots, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a grey hoodie, black track pants with white stripes, and grey sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair wearing a navy jacket, black trousers, and black sneakers, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 2 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 3 metres along its route; armed person 3 moves about 3 metres along its route; the protagonist moves about 4 metres along its route. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical noise of the industrial facility.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..720f6d61aa7a3b615a8640c0af55e12397353fea
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in an olive jacket stands in a paved industrial yard facing three armed adversaries, with warehouses and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a person seen from behind wearing an olive jacket, black trousers, and brown boots, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked concrete tarmac surface marked with mustard painted lines. The background features large corrugated metal warehouses with roll-up doors, a tall concrete observation tower, and stacked shipping containers in rust and navy hues. Distant mountains are visible under a bright, overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair wearing a brown leather jacket, khaki trousers, and black boots, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a grey hoodie, black track pants with white stripes, and grey sneakers, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair wearing a navy jacket, black trousers, and black sneakers, holding a handgun with both hands; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked concrete tarmac surface marked with mustard painted lines. The background features large corrugated metal warehouses with roll-up doors, a tall concrete observation tower, and stacked shipping containers in rust and navy hues. Distant mountains are visible under a bright, overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a person seen from behind wearing an olive jacket, black trousers, and brown boots, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair wearing a brown leather jacket, khaki trousers, and black boots, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a grey hoodie, black track pants with white stripes, and grey sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair wearing a navy jacket, black trousers, and black sneakers, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 2 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 3 metres along its route; armed person 3 moves about 3 metres along its route; the protagonist moves about 4 metres along its route. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical noise of the industrial facility.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..92e2682462cec19f7b92f83455bac650377fbe5e
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in an olive jacket stands in a paved industrial yard facing three armed adversaries, with warehouses and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a person seen from behind wearing an olive jacket, black trousers, and brown boots, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked concrete tarmac surface marked with mustard painted lines. The background features large corrugated metal warehouses with roll-up doors, a tall concrete observation tower, and stacked shipping containers in rust and navy hues. Distant mountains are visible under a bright, overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair wearing a brown leather jacket, khaki trousers, and black boots, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a grey hoodie, black track pants with white stripes, and grey sneakers, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair wearing a navy jacket, black trousers, and black sneakers, holding a handgun with both hands; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked concrete tarmac surface marked with mustard painted lines. The background features large corrugated metal warehouses with roll-up doors, a tall concrete observation tower, and stacked shipping containers in rust and navy hues. Distant mountains are visible under a bright, overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a person seen from behind wearing an olive jacket, black trousers, and brown boots, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair wearing a brown leather jacket, khaki trousers, and black boots, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a grey hoodie, black track pants with white stripes, and grey sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair wearing a navy jacket, black trousers, and black sneakers, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 2 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 3 metres along its route; armed person 3 moves about 3 metres along its route; the protagonist moves about 4 metres along its route. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical noise of the industrial facility.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a84acdf8abe34ec2635250a4219e3b52e633cc67
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands in a paved industrial yard facing three armed adversaries holding handguns, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan boots, standing in a ready stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked concrete tarmac surface marked with mustard painted lines. The background features large industrial warehouses with corrugated metal siding and roll-up doors, a tall concrete observation tower, stacked shipping containers in rust and navy blue, and distant mountain ridges under a bright overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a male figure in the middle distance wearing a grey hoodie, black trousers, and white sneakers, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a male figure in the middle distance wearing a moss-toned jacket over a white shirt, blue jeans, and white sneakers, holding a handgun; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+<Subject 5> is armed person 3: in <Picture 1> a male figure to the left wearing a brown leather jacket, khaki trousers, and dark boots, holding a handgun; in <Video 1> the human proxy tagged #cdf47b that starts on the left of the frame, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked concrete tarmac surface marked with mustard painted lines. The background features large industrial warehouses with corrugated metal siding and roll-up doors, a tall concrete observation tower, stacked shipping containers in rust and navy blue, and distant mountain ridges under a bright overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan boots, standing in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a male figure in the middle distance wearing a grey hoodie, black trousers, and white sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a male figure in the middle distance wearing a moss-toned jacket over a white shirt, blue jeans, and white sneakers, holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, is a male figure to the left wearing a brown leather jacket, khaki trousers, and dark boots, holding a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 2 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 3 metres along its route; armed person 3 moves about 3 metres along its route. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open yard and the distant clatter of industrial machinery.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..98601fdeb22e4abeaad112cd860bb4672cfbe13a
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands in a paved industrial yard facing three armed adversaries holding handguns, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan boots, standing in a ready stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked concrete tarmac surface marked with mustard painted lines. The background features large industrial warehouses with corrugated metal siding and roll-up doors, a tall concrete observation tower, stacked shipping containers in rust and navy blue, and distant mountain ridges under a bright overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a male figure in the middle distance wearing a grey hoodie, black trousers, and white sneakers, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a male figure in the middle distance wearing a moss-toned jacket over a white shirt, blue jeans, and white sneakers, holding a handgun; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: in <Picture 1> a male figure to the left wearing a brown leather jacket, khaki trousers, and dark boots, holding a handgun; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked concrete tarmac surface marked with mustard painted lines. The background features large industrial warehouses with corrugated metal siding and roll-up doors, a tall concrete observation tower, stacked shipping containers in rust and navy blue, and distant mountain ridges under a bright overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan boots, standing in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a male figure in the middle distance wearing a grey hoodie, black trousers, and white sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a male figure in the middle distance wearing a moss-toned jacket over a white shirt, blue jeans, and white sneakers, holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, is a male figure to the left wearing a brown leather jacket, khaki trousers, and dark boots, holding a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 2 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 3 metres along its route; armed person 3 moves about 3 metres along its route. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open yard and the distant clatter of industrial machinery.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8e138933a7e81a193a002347e07238f5e914915d
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-static_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,40 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a protagonist in tactical gear stands in a paved industrial yard facing three armed adversaries holding handguns, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 4 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan boots, standing in a ready stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked concrete tarmac surface marked with mustard painted lines. The background features large industrial warehouses with corrugated metal siding and roll-up doors, a tall concrete observation tower, stacked shipping containers in rust and navy blue, and distant mountain ridges under a bright overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a male figure in the middle distance wearing a grey hoodie, black trousers, and white sneakers, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a male figure in the middle distance wearing a moss-toned jacket over a white shirt, blue jeans, and white sneakers, holding a handgun; in <Video 1> the grey block-figure proxy.
+<Subject 5> is armed person 3: in <Picture 1> a male figure to the left wearing a brown leather jacket, khaki trousers, and dark boots, holding a handgun; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4> and <Subject 5>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked concrete tarmac surface marked with mustard painted lines. The background features large industrial warehouses with corrugated metal siding and roll-up doors, a tall concrete observation tower, stacked shipping containers in rust and navy blue, and distant mountain ridges under a bright overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan boots, standing in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a male figure in the middle distance wearing a grey hoodie, black trousers, and white sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a male figure in the middle distance wearing a moss-toned jacket over a white shirt, blue jeans, and white sneakers, holding a handgun, as seen in <Picture 1>. <Subject 5>, armed person 3, is a male figure to the left wearing a brown leather jacket, khaki trousers, and dark boots, holding a handgun, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they move slowly across the open ground and fire back; armed person 1 moves about 4 metres along its route; armed person 2 moves about 2 metres along its route; armed person 3 moves about 2 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 3 metres along its route; armed person 3 moves about 3 metres along its route. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open yard and the distant clatter of industrial machinery.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-05-cam-static_obj-enter-death/follow/mixed/prompt.txt b/prompts/neon/ndm-05-cam-static_obj-enter-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a2e22f64240d64bd304f608cd9508afd00f60116
--- /dev/null
+++ b/prompts/neon/ndm-05-cam-static_obj-enter-death/follow/mixed/prompt.txt
@@ -0,0 +1,42 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a tactical standoff in a paved industrial yard where a protagonist in a black vest faces three armed men near warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 5 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2), <Subject 5> (armed person 3), <Subject 6> (armed person 4); nothing else in <Video 1> is a person or an interaction object. <Subject 6> is not in frame at 00:00.000 and enters at about 1.9 s.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots. He stands in a ready stance facing the group ahead; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a vast, cracked asphalt tarmac marked with mustard painted lines. The background features large corrugated metal industrial warehouses with roll-up doors, a central loading gantry, and a tall concrete observation tower. To the right, stacks of blue and rust-coloured shipping containers sit near a building. Distant grey mountain ridges rise against a pale, overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair wearing a grey hoodie, black track pants with white side stripes, and white sneakers. He stands in the middle distance holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a sage jacket over a grey shirt, blue jeans, and white sneakers. He stands to the right, aiming a handgun forward; in <Video 1> the blue mannequin proxy.
+<Subject 5> is armed person 3: in <Picture 1> a man with dark hair wearing a maroon t-shirt, khaki trousers, and dark shoes. He stands further back to the left, holding a handgun; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 6> is armed person 4: appearance is not established by <Picture 1>; it enters the frame at about 1.9 s; in <Video 1> the blue mannequin proxy that starts on the left of the frame; preserve its recorded identity without inferring unseen clothing or materials from the image.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 4 down in an armed standoff as a newcomer walks in and the others fire back, involving <Subject 3> and <Subject 4> and <Subject 5> and <Subject 6>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 5> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 6> (enters [Shot 1] at about 1.9 s): partially_preserved - retain identity, geometry, trajectory and interaction timing from <Video 1>; appearance is not established by <Picture 1>, so do not claim to preserve unseen details from that image.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a vast, cracked asphalt tarmac marked with mustard painted lines. The background features large corrugated metal industrial warehouses with roll-up doors, a central loading gantry, and a tall concrete observation tower. To the right, stacks of blue and rust-coloured shipping containers sit near a building. Distant grey mountain ridges rise against a pale, overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots. He stands in a ready stance facing the group ahead, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair wearing a grey hoodie, black track pants with white side stripes, and white sneakers. He stands in the middle distance holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a sage jacket over a grey shirt, blue jeans, and white sneakers. He stands to the right, aiming a handgun forward, as seen in <Picture 1>. <Subject 5>, armed person 3, is a man with dark hair wearing a maroon t-shirt, khaki trousers, and dark shoes. He stands further back to the left, holding a handgun, as seen in <Picture 1>. <Subject 6>, armed person 4, has no discernible appearance in <Picture 1>; preserve its recorded identity and motion from <Video 1> without treating unspecified appearance as image evidence. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:03.267, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they fire back while another armed person walks into view; armed person 4 moves about 5 metres along its route. From 00:03.267 to 00:05.000, the protagonist shoots armed person 4, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 4 drops about half a metre. Keep the reference's fixed 51.246-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of wind across the open tarmac and the distant clatter of metal containers, punctuated by the sharp mechanical clicks of firearms being cocked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d847e7e0bece4668e0a6f79cb23dcbc77f2a4229
--- /dev/null
+++ b/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald protagonist in a tactical vest stands on an airport tarmac facing two armed men who are aiming weapons at him, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a bald man viewed from behind, wearing a grey tactical vest over a black long-sleeved shirt, camouflage trousers and black boots, standing with arms slightly raised; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with taxiway markings, hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky. Realize every recorded structure, vehicle and container with plausible materials appropriate to this setting; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short brown hair and a beard, wearing a brown leather jacket, khaki trousers and dark shoes, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a navy blazer over a white shirt, dark trousers and grey sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with taxiway markings, hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky. Realize every recorded structure, vehicle and container with plausible materials appropriate to this setting, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald man viewed from behind, wearing a grey tactical vest over a black long-sleeved shirt, camouflage trousers and black boots, standing with arms slightly raised, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short brown hair and a beard, wearing a brown leather jacket, khaki trousers and dark shoes, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a navy blazer over a white shirt, dark trousers and grey sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 5 metres along its route; armed person 2 moves about 3 metres along its route, raises the hands to chest height. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.809-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..cfae5accb60007c18e8b35f3967f0c559d42c901
--- /dev/null
+++ b/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald protagonist in a tactical vest stands on an airport tarmac facing two armed men who are aiming weapons at him, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a bald man viewed from behind, wearing a grey tactical vest over a black long-sleeved shirt, camouflage trousers and black boots, standing with arms slightly raised; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with taxiway markings, hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky. Realize every recorded structure, vehicle and container with plausible materials appropriate to this setting; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short brown hair and a beard, wearing a brown leather jacket, khaki trousers and dark shoes, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a navy blazer over a white shirt, dark trousers and grey sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with taxiway markings, hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky. Realize every recorded structure, vehicle and container with plausible materials appropriate to this setting, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald man viewed from behind, wearing a grey tactical vest over a black long-sleeved shirt, camouflage trousers and black boots, standing with arms slightly raised, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short brown hair and a beard, wearing a brown leather jacket, khaki trousers and dark shoes, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a navy blazer over a white shirt, dark trousers and grey sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 5 metres along its route; armed person 2 moves about 3 metres along its route, raises the hands to chest height. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.809-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d83c4ebdb4b133ac9427d0e1355177dd69c36e68
--- /dev/null
+++ b/prompts/neon/ndm-06-cam-static_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald protagonist in a tactical vest stands on an airport tarmac facing two armed men who are aiming weapons at him, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a bald man viewed from behind, wearing a grey tactical vest over a black long-sleeved shirt, camouflage trousers and black boots, standing with arms slightly raised; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with taxiway markings, hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky. Realize every recorded structure, vehicle and container with plausible materials appropriate to this setting; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with short brown hair and a beard, wearing a brown leather jacket, khaki trousers and dark shoes, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a navy blazer over a white shirt, dark trousers and grey sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with taxiway markings, hangars with segmented doors, a control tower with a glazed observation cabin, parked passenger aircraft, stacked cargo containers and floodlight masts, with distant mountain ridges under an overcast sky. Realize every recorded structure, vehicle and container with plausible materials appropriate to this setting, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald man viewed from behind, wearing a grey tactical vest over a black long-sleeved shirt, camouflage trousers and black boots, standing with arms slightly raised, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short brown hair and a beard, wearing a brown leather jacket, khaki trousers and dark shoes, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a navy blazer over a white shirt, dark trousers and grey sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 5 metres along its route; armed person 2 moves about 3 metres along its route, raises the hands to chest height. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.809-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The low hum of distant aircraft engines and wind across the tarmac, punctuated by the sharp metallic click of weapon safeties being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-06-cam-static_obj-static-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-06-cam-static_obj-static-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ac1c0a134a107fbb3ea2719b8f1ab941b4ac2abd
--- /dev/null
+++ b/prompts/neon/ndm-06-cam-static_obj-static-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a concrete airport apron, facing each other with weapons drawn, surrounded by large industrial hangars and distant mountains, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with a shaved head, viewed from behind. He wears a black tactical vest over a dark shirt, olive cargo trousers, and tan boots. He stands facing forward, aiming a handgun at the other two men; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features three large, grey corrugated metal hangars with segmented doors; the central hangar has a glazed control tower structure on its roof. Parked passenger aircraft are visible near the hangars, along with stacked cargo containers and tall floodlight masts. Distant mountain ridges rise under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair and a beard, wearing a navy blazer over a white shirt, black trousers, and black shoes. He stands in a three-quarter pose facing right, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a red plaid flannel shirt, grey jeans, and tan boots. He stands facing left, holding a handgun with both hands; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features three large, grey corrugated metal hangars with segmented doors; the central hangar has a glazed control tower structure on its roof. Parked passenger aircraft are visible near the hangars, along with stacked cargo containers and tall floodlight masts. Distant mountain ridges rise under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with a shaved head, viewed from behind. He wears a black tactical vest over a dark shirt, olive cargo trousers, and tan boots. He stands facing forward, aiming a handgun at the other two men, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair and a beard, wearing a navy blazer over a white shirt, black trousers, and black shoes. He stands in a three-quarter pose facing right, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a red plaid flannel shirt, grey jeans, and tan boots. He stands facing left, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.809-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and distant airport machinery, punctuated by the sharp metallic click of weapon slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-06-cam-static_obj-static-death/follow/mixed/prompt.txt b/prompts/neon/ndm-06-cam-static_obj-static-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3e1c12cbbfb14bbc842b63d032a20887db95b563
--- /dev/null
+++ b/prompts/neon/ndm-06-cam-static_obj-static-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a concrete airport apron, facing each other with weapons drawn, surrounded by large industrial hangars and distant mountains, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with a shaved head, viewed from behind. He wears a black tactical vest over a dark shirt, olive cargo trousers, and tan boots. He stands facing forward, aiming a handgun at the other two men; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features three large, grey corrugated metal hangars with segmented doors; the central hangar has a glazed control tower structure on its roof. Parked passenger aircraft are visible near the hangars, along with stacked cargo containers and tall floodlight masts. Distant mountain ridges rise under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair and a beard, wearing a navy blazer over a white shirt, black trousers, and black shoes. He stands in a three-quarter pose facing right, holding a handgun with both hands; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a red plaid flannel shirt, grey jeans, and tan boots. He stands facing left, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features three large, grey corrugated metal hangars with segmented doors; the central hangar has a glazed control tower structure on its roof. Parked passenger aircraft are visible near the hangars, along with stacked cargo containers and tall floodlight masts. Distant mountain ridges rise under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with a shaved head, viewed from behind. He wears a black tactical vest over a dark shirt, olive cargo trousers, and tan boots. He stands facing forward, aiming a handgun at the other two men, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair and a beard, wearing a navy blazer over a white shirt, black trousers, and black shoes. He stands in a three-quarter pose facing right, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a red plaid flannel shirt, grey jeans, and tan boots. He stands facing left, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.809-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and distant airport machinery, punctuated by the sharp metallic click of weapon slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-06-cam-static_obj-static-death/follow/white/prompt.txt b/prompts/neon/ndm-06-cam-static_obj-static-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..18b642dac7f9387116bcc1a189c0d4d02177971a
--- /dev/null
+++ b/prompts/neon/ndm-06-cam-static_obj-static-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a concrete airport apron, facing each other with weapons drawn, surrounded by large industrial hangars and distant mountains, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with a shaved head, viewed from behind. He wears a black tactical vest over a dark shirt, olive cargo trousers, and tan boots. He stands facing forward, aiming a handgun at the other two men; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features three large, grey corrugated metal hangars with segmented doors; the central hangar has a glazed control tower structure on its roof. Parked passenger aircraft are visible near the hangars, along with stacked cargo containers and tall floodlight masts. Distant mountain ridges rise under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair and a beard, wearing a navy blazer over a white shirt, black trousers, and black shoes. He stands in a three-quarter pose facing right, holding a handgun with both hands; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a red plaid flannel shirt, grey jeans, and tan boots. He stands facing left, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, cracked concrete tarmac marked with yellow taxiway lines. The background features three large, grey corrugated metal hangars with segmented doors; the central hangar has a glazed control tower structure on its roof. Parked passenger aircraft are visible near the hangars, along with stacked cargo containers and tall floodlight masts. Distant mountain ridges rise under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with a shaved head, viewed from behind. He wears a black tactical vest over a dark shirt, olive cargo trousers, and tan boots. He stands facing forward, aiming a handgun at the other two men, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair and a beard, wearing a navy blazer over a white shirt, black trousers, and black shoes. He stands in a three-quarter pose facing right, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a red plaid flannel shirt, grey jeans, and tan boots. He stands facing left, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.809-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and distant airport machinery, punctuated by the sharp metallic click of weapon slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2bd540d163a40722338155c582e03ee719c41fbd
--- /dev/null
+++ b/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a cracked concrete airport apron, facing each other with weapons drawn, surrounded by hangars and a control tower, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, viewed from behind, wearing an olive jacket and dark cargo trousers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, weathered concrete apron with visible cracks and weeds growing in the seams, marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked brown cargo containers, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a grey hooded sweatshirt and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with short hair, wearing a brown leather jacket and tan trousers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, weathered concrete apron with visible cracks and weeds growing in the seams, marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked brown cargo containers, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, viewed from behind, wearing an olive jacket and dark cargo trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a grey hooded sweatshirt and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short hair, wearing a brown leather jacket and tan trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about half a metre along its route, raises the hands to chest height; the protagonist moves about 4 metres along its route. Keep the reference's fixed 54.307-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical noise of airport operations, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..69776dabf78b2bcdab63b6c35b0d84f728ed7be6
--- /dev/null
+++ b/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a cracked concrete airport apron, facing each other with weapons drawn, surrounded by hangars and a control tower, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, viewed from behind, wearing an olive jacket and dark cargo trousers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, weathered concrete apron with visible cracks and weeds growing in the seams, marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked brown cargo containers, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a grey hooded sweatshirt and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short hair, wearing a brown leather jacket and tan trousers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, weathered concrete apron with visible cracks and weeds growing in the seams, marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked brown cargo containers, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, viewed from behind, wearing an olive jacket and dark cargo trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a grey hooded sweatshirt and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short hair, wearing a brown leather jacket and tan trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about half a metre along its route, raises the hands to chest height; the protagonist moves about 4 metres along its route. Keep the reference's fixed 54.307-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical noise of airport operations, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b58f7a04643d84e557ca036e3b55d29f44ed92c2
--- /dev/null
+++ b/prompts/neon/ndm-07-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a cracked concrete airport apron, facing each other with weapons drawn, surrounded by hangars and a control tower, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, viewed from behind, wearing an olive jacket and dark cargo trousers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, weathered concrete apron with visible cracks and weeds growing in the seams, marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked brown cargo containers, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a grey hooded sweatshirt and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short hair, wearing a brown leather jacket and tan trousers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, weathered concrete apron with visible cracks and weeds growing in the seams, marked with yellow taxiway lines. The background features large industrial hangars with segmented metal doors, a control tower with a glazed observation deck, stacked brown cargo containers, and a parked passenger aircraft. Distant mountain ridges are visible under an overcast sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, viewed from behind, wearing an olive jacket and dark cargo trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a grey hooded sweatshirt and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short hair, wearing a brown leather jacket and tan trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 2 metres along its route; armed person 2 moves about 2 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about half a metre along its route, raises the hands to chest height; the protagonist moves about 4 metres along its route. Keep the reference's fixed 54.307-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical noise of airport operations, punctuated by the sharp metallic click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2646d758e23f8ac4f976bf78246a8cbbadda1b3b
--- /dev/null
+++ b/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff formation in a paved industrial loading yard, facing each other with weapons drawn, surrounded by warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket, black cargo trousers, and brown boots. He is viewed from behind, aiming a handgun forward; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a paved asphalt surface marked with yellow parking lines and white directional arrows. The background features large industrial warehouses with corrugated metal siding and concrete bases. To the left, a stack of green and rust-red shipping containers sits against a building. In the center distance, a large open bay reveals an orange loading gantry. To the right, a warehouse with a closed roll-up door stands near a tall floodlight mast. Distant mountains are visible under a hazy sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair, wearing a grey zip-up hoodie, black track pants with white side stripes, and white sneakers. He stands in a defensive stance, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a brown leather jacket, khaki trousers, and brown boots. He stands aiming a handgun forward; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a paved asphalt surface marked with yellow parking lines and white directional arrows. The background features large industrial warehouses with corrugated metal siding and concrete bases. To the left, a stack of green and rust-red shipping containers sits against a building. In the center distance, a large open bay reveals an orange loading gantry. To the right, a warehouse with a closed roll-up door stands near a tall floodlight mast. Distant mountains are visible under a hazy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket, black cargo trousers, and brown boots. He is viewed from behind, aiming a handgun forward, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair, wearing a grey zip-up hoodie, black track pants with white side stripes, and white sneakers. He stands in a defensive stance, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a brown leather jacket, khaki trousers, and brown boots. He stands aiming a handgun forward, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 2 moves about 2 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 2 metres along its route. Keep the reference's fixed 50.984-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind in the open yard and distant traffic, punctuated by the sharp metallic click of a pistol slide being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9e72c83d420a6316aca2961455ce0f36ebb0d5ac
--- /dev/null
+++ b/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff formation in a paved industrial loading yard, facing each other with weapons drawn, surrounded by warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket, black cargo trousers, and brown boots. He is viewed from behind, aiming a handgun forward; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a paved asphalt surface marked with yellow parking lines and white directional arrows. The background features large industrial warehouses with corrugated metal siding and concrete bases. To the left, a stack of green and rust-red shipping containers sits against a building. In the center distance, a large open bay reveals an orange loading gantry. To the right, a warehouse with a closed roll-up door stands near a tall floodlight mast. Distant mountains are visible under a hazy sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair, wearing a grey zip-up hoodie, black track pants with white side stripes, and white sneakers. He stands in a defensive stance, holding a handgun with both hands; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a brown leather jacket, khaki trousers, and brown boots. He stands aiming a handgun forward; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a paved asphalt surface marked with yellow parking lines and white directional arrows. The background features large industrial warehouses with corrugated metal siding and concrete bases. To the left, a stack of green and rust-red shipping containers sits against a building. In the center distance, a large open bay reveals an orange loading gantry. To the right, a warehouse with a closed roll-up door stands near a tall floodlight mast. Distant mountains are visible under a hazy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket, black cargo trousers, and brown boots. He is viewed from behind, aiming a handgun forward, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair, wearing a grey zip-up hoodie, black track pants with white side stripes, and white sneakers. He stands in a defensive stance, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a brown leather jacket, khaki trousers, and brown boots. He stands aiming a handgun forward, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 2 moves about 2 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 2 metres along its route. Keep the reference's fixed 50.984-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind in the open yard and distant traffic, punctuated by the sharp metallic click of a pistol slide being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4cfdb5c4dfc0996c723ad0c99c87e066dbba3a07
--- /dev/null
+++ b/prompts/neon/ndm-08-cam-static_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff formation in a paved industrial loading yard, facing each other with weapons drawn, surrounded by warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket, black cargo trousers, and brown boots. He is viewed from behind, aiming a handgun forward; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a paved asphalt surface marked with yellow parking lines and white directional arrows. The background features large industrial warehouses with corrugated metal siding and concrete bases. To the left, a stack of green and rust-red shipping containers sits against a building. In the center distance, a large open bay reveals an orange loading gantry. To the right, a warehouse with a closed roll-up door stands near a tall floodlight mast. Distant mountains are visible under a hazy sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair, wearing a grey zip-up hoodie, black track pants with white side stripes, and white sneakers. He stands in a defensive stance, holding a handgun with both hands; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a brown leather jacket, khaki trousers, and brown boots. He stands aiming a handgun forward; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a paved asphalt surface marked with yellow parking lines and white directional arrows. The background features large industrial warehouses with corrugated metal siding and concrete bases. To the left, a stack of green and rust-red shipping containers sits against a building. In the center distance, a large open bay reveals an orange loading gantry. To the right, a warehouse with a closed roll-up door stands near a tall floodlight mast. Distant mountains are visible under a hazy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket, black cargo trousers, and brown boots. He is viewed from behind, aiming a handgun forward, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair, wearing a grey zip-up hoodie, black track pants with white side stripes, and white sneakers. He stands in a defensive stance, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a brown leather jacket, khaki trousers, and brown boots. He stands aiming a handgun forward, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 2 moves about 2 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 2 metres along its route. Keep the reference's fixed 50.984-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind in the open yard and distant traffic, punctuated by the sharp metallic click of a pistol slide being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-08-cam-static_obj-static-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-08-cam-static_obj-static-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..80ce403707acd12c1ccb4a4b53aee7ac2b59753f
--- /dev/null
+++ b/prompts/neon/ndm-08-cam-static_obj-static-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a standoff in a paved industrial yard where a protagonist in tactical gear faces two armed men near warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with a shaved head, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked asphalt surface marked with mustard painted lines, surrounded by large corrugated metal warehouses with roll-up doors and small windows. A tall metal floodlight mast stands centrally, with a loading gantry visible in the background. Stacked shipping containers in rust and dark blue sit to the right, and distant mountains are visible under a hazy sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair and a beard, wearing a sage jacket over a white shirt, dark denim jeans, and white sneakers, holding a handgun with both hands; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a light grey hoodie, black track pants with a white stripe, and grey sneakers, holding a handgun with both hands; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked asphalt surface marked with mustard painted lines, surrounded by large corrugated metal warehouses with roll-up doors and small windows. A tall metal floodlight mast stands centrally, with a loading gantry visible in the background. Stacked shipping containers in rust and dark blue sit to the right, and distant mountains are visible under a hazy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with a shaved head, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots, holding a handgun with both hands, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair and a beard, wearing a sage jacket over a white shirt, dark denim jeans, and white sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a light grey hoodie, black track pants with a white stripe, and grey sneakers, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.984-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind in the open yard and the distant clatter of metal, punctuated by the sharp mechanical click of pistol slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-08-cam-static_obj-static-death/follow/mixed/prompt.txt b/prompts/neon/ndm-08-cam-static_obj-static-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b4f0e95466ffb627ef7244c025ea3b3ba1835a14
--- /dev/null
+++ b/prompts/neon/ndm-08-cam-static_obj-static-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a standoff in a paved industrial yard where a protagonist in tactical gear faces two armed men near warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with a shaved head, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked asphalt surface marked with mustard painted lines, surrounded by large corrugated metal warehouses with roll-up doors and small windows. A tall metal floodlight mast stands centrally, with a loading gantry visible in the background. Stacked shipping containers in rust and dark blue sit to the right, and distant mountains are visible under a hazy sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair and a beard, wearing a sage jacket over a white shirt, dark denim jeans, and white sneakers, holding a handgun with both hands; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a light grey hoodie, black track pants with a white stripe, and grey sneakers, holding a handgun with both hands; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked asphalt surface marked with mustard painted lines, surrounded by large corrugated metal warehouses with roll-up doors and small windows. A tall metal floodlight mast stands centrally, with a loading gantry visible in the background. Stacked shipping containers in rust and dark blue sit to the right, and distant mountains are visible under a hazy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with a shaved head, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots, holding a handgun with both hands, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair and a beard, wearing a sage jacket over a white shirt, dark denim jeans, and white sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a light grey hoodie, black track pants with a white stripe, and grey sneakers, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.984-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind in the open yard and the distant clatter of metal, punctuated by the sharp mechanical click of pistol slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-08-cam-static_obj-static-death/follow/white/prompt.txt b/prompts/neon/ndm-08-cam-static_obj-static-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f584b72361cca53849c5b2fedce2497dde6f6f02
--- /dev/null
+++ b/prompts/neon/ndm-08-cam-static_obj-static-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a standoff in a paved industrial yard where a protagonist in tactical gear faces two armed men near warehouses and cargo containers, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with a shaved head, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a cracked asphalt surface marked with mustard painted lines, surrounded by large corrugated metal warehouses with roll-up doors and small windows. A tall metal floodlight mast stands centrally, with a loading gantry visible in the background. Stacked shipping containers in rust and dark blue sit to the right, and distant mountains are visible under a hazy sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with short dark hair and a beard, wearing a sage jacket over a white shirt, dark denim jeans, and white sneakers, holding a handgun with both hands; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with short dark hair, wearing a light grey hoodie, black track pants with a white stripe, and grey sneakers, holding a handgun with both hands; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down in an armed standoff while the others fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a cracked asphalt surface marked with mustard painted lines, surrounded by large corrugated metal warehouses with roll-up doors and small windows. A tall metal floodlight mast stands centrally, with a loading gantry visible in the background. Stacked shipping containers in rust and dark blue sit to the right, and distant mountains are visible under a hazy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with a shaved head, wearing a black tactical vest over a dark shirt, olive cargo trousers, and tan work boots, holding a handgun with both hands, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with short dark hair and a beard, wearing a sage jacket over a white shirt, dark denim jeans, and white sneakers, holding a handgun with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with short dark hair, wearing a light grey hoodie, black track pants with a white stripe, and grey sneakers, holding a handgun with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; they hold their positions and fire back. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre. Keep the reference's fixed 50.984-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind in the open yard and the distant clatter of metal, punctuated by the sharp mechanical click of pistol slides being racked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..67dd5c46e0bc2372b975ffbf729b9f8ffee457b5
--- /dev/null
+++ b/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a concrete airport apron, aiming handguns at one another in front of large industrial hangars, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket and black cargo trousers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by large beige corrugated metal hangars with segmented sliding doors and a central building with a row of windows. A tall floodlight mast stands between the buildings. In the background, distant mountain ridges are visible under a sunset sky with soft clouds. A passenger aircraft is partially visible on the far right; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a grey hoodie and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47b7b that starts on the left of the frame, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a brown leather jacket and khaki trousers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #7baaf4, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by large beige corrugated metal hangars with segmented sliding doors and a central building with a row of windows. A tall floodlight mast stands between the buildings. In the background, distant mountain ridges are visible under a sunset sky with soft clouds. A passenger aircraft is partially visible on the far right, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket and black cargo trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a grey hoodie and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a brown leather jacket and khaki trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 51.784-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical whine of an aircraft engine, punctuated by the sharp metallic click of handgun hammers being cocked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e041b5de9e3c789853dd2b5be49d74e1cb7c2439
--- /dev/null
+++ b/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a concrete airport apron, aiming handguns at one another in front of large industrial hangars, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket and black cargo trousers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by large beige corrugated metal hangars with segmented sliding doors and a central building with a row of windows. A tall floodlight mast stands between the buildings. In the background, distant mountain ridges are visible under a sunset sky with soft clouds. A passenger aircraft is partially visible on the far right; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a grey hoodie and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a brown leather jacket and khaki trousers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by large beige corrugated metal hangars with segmented sliding doors and a central building with a row of windows. A tall floodlight mast stands between the buildings. In the background, distant mountain ridges are visible under a sunset sky with soft clouds. A passenger aircraft is partially visible on the far right, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket and black cargo trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a grey hoodie and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a brown leather jacket and khaki trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 51.784-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical whine of an aircraft engine, punctuated by the sharp metallic click of handgun hammers being cocked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..21c3021ed4a19db4aadc239262942864f825b288
--- /dev/null
+++ b/prompts/neon/ndm-09-cam-dynamic_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed men stand in a standoff on a concrete airport apron, aiming handguns at one another in front of large industrial hangars, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a man with short dark hair, wearing an olive jacket and black cargo trousers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by large beige corrugated metal hangars with segmented sliding doors and a central building with a row of windows. A tall floodlight mast stands between the buildings. In the background, distant mountain ridges are visible under a sunset sky with soft clouds. A passenger aircraft is partially visible on the far right; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard, wearing a grey hoodie and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy that starts on the left of the frame.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair and a beard, wearing a brown leather jacket and khaki trousers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings and tire tracks, flanked by large beige corrugated metal hangars with segmented sliding doors and a central building with a row of windows. A tall floodlight mast stands between the buildings. In the background, distant mountain ridges are visible under a sunset sky with soft clouds. A passenger aircraft is partially visible on the far right, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a man with short dark hair, wearing an olive jacket and black cargo trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard, wearing a grey hoodie and black track pants with white stripes down the sides, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair and a beard, wearing a brown leather jacket and khaki trousers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; the protagonist moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; the protagonist moves about 4 metres along its route. Keep the reference's fixed 51.784-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The ambient hum of wind across the open tarmac and the distant mechanical whine of an aircraft engine, punctuated by the sharp metallic click of handgun hammers being cocked.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/colored-obb/prompt.txt b/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..cbc7a21ef45e51a1a5d27a3694f450b65fb5bc92
--- /dev/null
+++ b/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff formation on a wet, paved airport apron, facing each other with weapons drawn, surrounded by large industrial hangars under a rainy sky, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing an olive jacket, black trousers, and brown boots, holding a firearm with both hands in a ready stance; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, grey concrete airport apron marked with yellow taxiway lines, featuring visible puddles reflecting the overcast sky. The background consists of large, industrial hangars with segmented metal doors and small rectangular windows. Tall floodlight masts stand between the buildings, and distant mountain ridges are visible through the mist and rain; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a male figure facing the camera, wearing a grey hoodie, black track pants with a white stripe, and grey sneakers, holding a firearm with both hands; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a male figure facing the camera, wearing a brown jacket, khaki trousers, and black shoes, holding a firearm with both hands; in <Video 1> the human proxy tagged #7baaf4 that starts on the right of the frame, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, grey concrete airport apron marked with yellow taxiway lines, featuring visible puddles reflecting the overcast sky. The background consists of large, industrial hangars with segmented metal doors and small rectangular windows. Tall floodlight masts stand between the buildings, and distant mountain ridges are visible through the mist and rain, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing an olive jacket, black trousers, and brown boots, holding a firearm with both hands in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a male figure facing the camera, wearing a grey hoodie, black track pants with a white stripe, and grey sneakers, holding a firearm with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a male figure facing the camera, wearing a brown jacket, khaki trousers, and black shoes, holding a firearm with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; armed person 2 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 4 metres along its route. Keep the reference's fixed 51.225-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The sound of steady rain falling on concrete and the distant rumble of thunder, punctuated by the mechanical click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/mixed/prompt.txt b/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f031b70f229cdb13f54ae8dfc770ef8236903ed7
--- /dev/null
+++ b/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff formation on a wet, paved airport apron, facing each other with weapons drawn, surrounded by large industrial hangars under a rainy sky, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing an olive jacket, black trousers, and brown boots, holding a firearm with both hands in a ready stance; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, grey concrete airport apron marked with yellow taxiway lines, featuring visible puddles reflecting the overcast sky. The background consists of large, industrial hangars with segmented metal doors and small rectangular windows. Tall floodlight masts stand between the buildings, and distant mountain ridges are visible through the mist and rain; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a male figure facing the camera, wearing a grey hoodie, black track pants with a white stripe, and grey sneakers, holding a firearm with both hands; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a male figure facing the camera, wearing a brown jacket, khaki trousers, and black shoes, holding a firearm with both hands; in <Video 1> the blue mannequin proxy that starts on the right of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, grey concrete airport apron marked with yellow taxiway lines, featuring visible puddles reflecting the overcast sky. The background consists of large, industrial hangars with segmented metal doors and small rectangular windows. Tall floodlight masts stand between the buildings, and distant mountain ridges are visible through the mist and rain, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing an olive jacket, black trousers, and brown boots, holding a firearm with both hands in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a male figure facing the camera, wearing a grey hoodie, black track pants with a white stripe, and grey sneakers, holding a firearm with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a male figure facing the camera, wearing a brown jacket, khaki trousers, and black shoes, holding a firearm with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; armed person 2 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 4 metres along its route. Keep the reference's fixed 51.225-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The sound of steady rain falling on concrete and the distant rumble of thunder, punctuated by the mechanical click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/white/prompt.txt b/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8f839dc0e3a8124fed757c956cd5e411fb87be58
--- /dev/null
+++ b/prompts/neon/ndm-10-cam-dynamic_obj-dynamic-death-composite-forward-back-v1/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing three armed adults stand in a standoff formation on a wet, paved airport apron, facing each other with weapons drawn, surrounded by large industrial hangars under a rainy sky, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a male figure seen from behind, wearing an olive jacket, black trousers, and brown boots, holding a firearm with both hands in a ready stance; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide, grey concrete airport apron marked with yellow taxiway lines, featuring visible puddles reflecting the overcast sky. The background consists of large, industrial hangars with segmented metal doors and small rectangular windows. Tall floodlight masts stand between the buildings, and distant mountain ridges are visible through the mist and rain; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a male figure facing the camera, wearing a grey hoodie, black track pants with a white stripe, and grey sneakers, holding a firearm with both hands; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a male figure facing the camera, wearing a brown jacket, khaki trousers, and black shoes, holding a firearm with both hands; in <Video 1> the grey block-figure proxy that starts on the right of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide, grey concrete airport apron marked with yellow taxiway lines, featuring visible puddles reflecting the overcast sky. The background consists of large, industrial hangars with segmented metal doors and small rectangular windows. Tall floodlight masts stand between the buildings, and distant mountain ridges are visible through the mist and rain, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a male figure seen from behind, wearing an olive jacket, black trousers, and brown boots, holding a firearm with both hands in a ready stance, as seen in <Picture 1>. <Subject 3>, armed person 1, is a male figure facing the camera, wearing a grey hoodie, black track pants with a white stripe, and grey sneakers, holding a firearm with both hands, as seen in <Picture 1>. <Subject 4>, armed person 2, is a male figure facing the camera, wearing a brown jacket, khaki trousers, and black shoes, holding a firearm with both hands, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; armed person 2 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 4 metres along its route. Keep the reference's fixed 51.225-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The sound of steady rain falling on concrete and the distant rumble of thunder, punctuated by the mechanical click of weapon safety catches being disengaged.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt b/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a49e1768a0add7981b5a0995c88cda25c4b7c623
--- /dev/null
+++ b/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/colored-obb/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, occupied scale (in this representation the box is the entity's exact extent), and whole-body root orientation, which are exact and must be followed precisely. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction. On every human, vehicle and animal proxy in <Video 1> the white face is the front of that entity's body: white there is a direction code, never an object, a material or a light, and it must not be painted. The hex colour tags used below only say which proxy stands for which subject: they are identity codes, never colours or materials of the result. Wherever <Video 1> shows a subject only as a flat single-colour shape, that shape is likewise only a placeholder for position, size and motion: draw the complete real person, animal or object the text names in its place.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald protagonist in tactical gear stands on a wet airport apron facing two armed men, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 human proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object. The proxy tagged #cdf47b (person) never enters the frame of this shot: no figure or object appears for it and none must be added.
+<Subject 1> is the protagonist: in <Picture 1> a bald adult male wearing a grey camouflage tactical vest over a black long-sleeved shirt, grey camouflage trousers and black boots, standing with his back to the camera; in <Video 1> the human proxy tagged #f47be9, whose white face is its front.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings, wet concrete ground reflecting the sky, large grey hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers in navy, emerald and amber, floodlight masts, and distant mountain ridges under a rainy sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard wearing a black t-shirt, dark trousers and black shoes, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #f47b7b, whose white face is its front.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a dark moss-toned jacket, blue jeans and white sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the human proxy tagged #7baaf4 that starts on the right of the frame, whose white face is its front.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation (the white face of the box is the front of the body), identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings, wet concrete ground reflecting the sky, large grey hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers in navy, emerald and amber, floodlight masts, and distant mountain ridges under a rainy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald adult male wearing a grey camouflage tactical vest over a black long-sleeved shirt, grey camouflage trousers and black boots, standing with his back to the camera, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard wearing a black t-shirt, dark trousers and black shoes, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a dark moss-toned jacket, blue jeans and white sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; armed person 2 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 4 metres along its route. Keep the reference's fixed 51.225-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The sound of rain falling on concrete and distant thunder, accompanied by the heavy breathing of the characters and the mechanical click of a weapon being readied.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/mixed/prompt.txt b/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fa2b749fc35b16f37fd5df8cc76ad74fca6247d4
--- /dev/null
+++ b/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/mixed/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide only framewise root position, coarse occupied scale, and whole-body root orientation, which should be followed closely. The proxy figure keeps ONE fixed upright pose for the whole clip: that pose is not an action and carries no information about sitting, crouching, riding, reaching or any limb articulation. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors, visual effects, anatomy, bones, joints, limb articulation, detailed posture, torso or head pose, hand pose, action phase, or physical contact. Reconstruct each character's appearance from <Picture 1> and the text, and reconstruct articulated motion from the timestamped text. Never reproduce a human proxy as a skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald protagonist in tactical gear stands on a wet airport apron facing two armed men, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 blue mannequin proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a bald adult male wearing a grey camouflage tactical vest over a black long-sleeved shirt, grey camouflage trousers and black boots, standing with his back to the camera; in <Video 1> the blue mannequin proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings, wet concrete ground reflecting the sky, large grey hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers in navy, emerald and amber, floodlight masts, and distant mountain ridges under a rainy sky; in <Video 1> the same environment layout is represented by flat semantic-colour proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials or conditioning colours.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard wearing a black t-shirt, dark trousers and black shoes, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a dark moss-toned jacket, blue jeans and white sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the blue mannequin proxy that starts on the right of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, limb articulation, detailed posture and physical contact are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and its articulated motion from the timeline below.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings, wet concrete ground reflecting the sky, large grey hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers in navy, emerald and amber, floodlight masts, and distant mountain ridges under a rainy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald adult male wearing a grey camouflage tactical vest over a black long-sleeved shirt, grey camouflage trousers and black boots, standing with his back to the camera, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard wearing a black t-shirt, dark trousers and black shoes, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a dark moss-toned jacket, blue jeans and white sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; armed person 2 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 4 metres along its route. Keep the reference's fixed 51.225-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The sound of rain falling on concrete and distant thunder, accompanied by the heavy breathing of the characters and the mechanical click of a weapon being readied.
+
+non_diegetic_music:
+N/A
diff --git a/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/white/prompt.txt b/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/white/prompt.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3da98a131bfa287a511db426cb60d8291075804d
--- /dev/null
+++ b/prompts/neon/ndm-10-cam-static_obj-dynamic-death/follow/white/prompt.txt
@@ -0,0 +1,38 @@
+AWM_PROXY_CONTROL. Two visual references accompany this text. <Video 1> is the conditioning proxy video: treat it as the exact spatiotemporal reference for where every subject is, how large it is and how it moves, but never as an appearance template or literal target geometry. <Picture 1> is the global RGB anchor image: it defines identity and appearance, never motion. <Picture 1> is the ABSOLUTE standard for the worldview, art direction and overall style of the entire take, and on worldview, art direction and style it outranks even the description below. It provides NOTHING about any later frame's camera, composition, motion or interactions. Each subject's position and motion come from <Video 1>. A subject's appearance follows <Picture 1> when that subject is visible in it; subjects not visible in <Picture 1> take their appearance from the text, rendered consistently in <Picture 1>'s worldview and art direction.
+
+Human proxies in <Video 1> provide framewise root position, occupied scale, whole-body root orientation and COARSE limb articulation: the proxy figure's limbs move with the recorded joints, so follow the timing and direction of its limb motion closely, and take everything finer - hand pose, facial expression, grip and physical contact - from the timestamped text. They carry no information about character appearance and do not represent identity, face, hairstyle, clothing, materials, colors or visual effects. Reconstruct each character's appearance from <Picture 1> and the text. Never reproduce a human proxy as a block figure, skeleton, stick figure, mannequin, node-connected body, or exposed rig. Left and right refer to each person's own body unless explicitly stated as screen direction.
+
+Roads, terrain, buildings, relative depth, rough scale, and layout in <Video 1> provide denser but still approximate structural cues. Conditioning colors, proxy geometry, node markers, rigging structures, and other visualization artifacts must never appear in the final video.
+
+When sources conflict, resolve in this order: (1) geometry, camera, position and motion follow <Video 1>; (2) the worldview, art direction and overall style follow <Picture 1> absolutely - above the description below; identity, appearance and materials follow <Picture 1> for subjects visible in it, and follow the text for subjects not visible in it; (3) the timestamped text defines the concrete events, actions and rewrite targets within those constraints.
+
+Clip contract: this clip is the very beginning of the take - nothing precedes it. <Picture 1> is the global appearance anchor for the whole take, and the first frame of the target is locked to this exact image: the clip begins at precisely this visual state - identities, appearance, materials, art direction, lighting, camera framing and layout - and continues directly from it while performing the timestamped actions. Do not restart, redesign or re-interpret any identity.
+
+subject_definitions:
+<Picture 1> is the first frame of [Shot 1]: the photoreal opening frame of the target video, showing a bald protagonist in tactical gear stands on a wet airport apron facing two armed men, with hangars and a control tower in the background, exactly as they must appear at 00:00.000.
+<Video 1> is the conditioning proxy video of the same 5-second shot (its role is fixed by the control block above). The clip contains exactly 3 grey block-figure proxies standing for <Subject 1> (the protagonist), <Subject 3> (armed person 1), <Subject 4> (armed person 2); nothing else in <Video 1> is a person or an interaction object.
+<Subject 1> is the protagonist: in <Picture 1> a bald adult male wearing a grey camouflage tactical vest over a black long-sleeved shirt, grey camouflage trousers and black boots, standing with his back to the camera; in <Video 1> the grey block-figure proxy.
+<Subject 2> is the environment and its fixed spatial arrangement: in <Picture 1> a wide paved airport apron with yellow taxiway markings, wet concrete ground reflecting the sky, large grey hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers in navy, emerald and amber, floodlight masts, and distant mountain ridges under a rainy sky; in <Video 1> the same environment layout is represented by untextured grey proxy geometry: approximate structural cues for layout, relative depth and rough scale, never target materials.
+<Subject 3> is armed person 1: in <Picture 1> a man with dark hair and a beard wearing a black t-shirt, dark trousers and black shoes, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy.
+<Subject 4> is armed person 2: in <Picture 1> a man with dark hair wearing a dark moss-toned jacket, blue jeans and white sneakers, holding a handgun with both hands in a firing stance; in <Video 1> the grey block-figure proxy that starts on the right of the frame.
+
+summary:
+[keyframe completion + reference generation] Starting exactly from <Picture 1> as the first frame of [Shot 1], generate a 5.00-second continuous shot in <Subject 2> of <Subject 1> shooting armed person 1 down while armed people on the move fire back, involving <Subject 3> and <Subject 4>. Use <Video 1> for layout, root motion, limb motion, object trajectories, action timing and the camera path, and keep the appearance, materials and lighting established in <Picture 1>.
+
+retention_analysis:
+<Picture 1> ([Shot 1] first frame): fully_preserved - the first frame of the target video reproduces this image exactly, including every person's appearance, the objects, the scene layout, materials, lighting and framing; later frames evolve continuously from it.
+<Video 1> (whole-shot spatial and temporal structure): fully_preserved - geometry, camera, placement, root motion, recorded limb motion and object trajectories follow <Video 1> as a coarse spatiotemporal reference; appearance, materials and surface detail are not encoded in the proxy.
+<Subject 1> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 2> (appears in [Shot 1]): partially_preserved - retain terrain, object layout and architectural forms from <Video 1>; surfaces, materials and lighting come from <Picture 1>.
+<Subject 3> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+<Subject 4> (appears in [Shot 1]): partially_preserved - retain the proxy's framewise root position, occupied scale, root orientation, recorded limb motion, identity continuity and interaction timing from <Video 1>; reconstruct its appearance from <Picture 1> and render natural human articulation over the recorded block-figure motion.
+
+detailed_description:
+Photorealistic natural lighting, physically plausible scale, detailed real-world materials and natural human anatomy; no proxy colours, wireframes, block figures or game UI.
+[Shot 1] The shot opens exactly on <Picture 1> and evolves continuously from that frame. <Subject 2>, the setting, is a wide paved airport apron with yellow taxiway markings, wet concrete ground reflecting the sky, large grey hangars with segmented doors, a control tower with a glazed observation cabin, stacked cargo containers in navy, emerald and amber, floodlight masts, and distant mountain ridges under a rainy sky, keeping the spacing and sizes of <Video 1>. <Subject 1>, the protagonist, is a bald adult male wearing a grey camouflage tactical vest over a black long-sleeved shirt, grey camouflage trousers and black boots, standing with his back to the camera, as seen in <Picture 1>. <Subject 3>, armed person 1, is a man with dark hair and a beard wearing a black t-shirt, dark trousers and black shoes, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. <Subject 4>, armed person 2, is a man with dark hair wearing a dark moss-toned jacket, blue jeans and white sneakers, holding a handgun with both hands in a firing stance, as seen in <Picture 1>. All people keep the same face, build and clothing throughout, with clothing folds responding naturally to movement. The action follows <Video 1> according to this timeline: From 00:00.000 to 00:02.767, the protagonist keeps a handgun raised in both hands, aimed at the armed people in front of them, and fires at them; some of them move slowly across the open ground and fire back; armed person 1 moves about 3 metres along its route; armed person 2 moves about 4 metres along its route. From 00:02.767 to 00:05.000, the protagonist shoots armed person 1, who collapses to the ground within about half a second, then lies motionless for the rest of the shot; everyone else stays on their feet, the protagonist keeps the handgun raised and the other armed people keep firing back; armed person 1 drops about half a metre; armed person 2 moves about 4 metres along its route. Keep the reference's fixed 51.225-degree vertical field of view. Preserve the single take and its pacing, without cuts or a new camera path. Keep world geometry, object identities and materials stable throughout the camera movement; reveal newly visible surfaces with the same layout and appearance; introduce no extra people or foreground obstacles. No interface overlays, captions, watermarks, dialogue or music.
+
+overall_soundscape:
+The sound of rain falling on concrete and distant thunder, accompanied by the heavy breathing of the characters and the mechanical click of a weapon being readied.
+
+non_diegetic_music:
+N/A
diff --git a/splits/full.parquet b/splits/full.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..42c42c916903a84c1b96b60a87f62fdd6c48c2b8
--- /dev/null
+++ b/splits/full.parquet
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d95aa0a8d0b3a13881cca979c0a391b545b19a37359063a1c9bb370997375e77
+size 304870
diff --git a/splits/long_video.parquet b/splits/long_video.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..8b0aba3baaff43c4654e72a26eb30e71600f7003
--- /dev/null
+++ b/splits/long_video.parquet
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e9b1cf87a5f8cafa36bcda63f849cd3a2d348c29b70ad6165276966ce6a216c
+size 28951
diff --git a/splits/multi_view.parquet b/splits/multi_view.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..4c990965e73c7bda8430870906409f5b3cadde2f
--- /dev/null
+++ b/splits/multi_view.parquet
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8c66dd0dfd80ca715d7dfdbd3e43824fd9be8cf3a95bab9b33208409f88f305
+size 48000
diff --git a/splits/neon.parquet b/splits/neon.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..145a21a5359170b6158c0c2460a72832e79fb39e
--- /dev/null
+++ b/splits/neon.parquet
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99ab9c5c7e67ab6a09597444f339307fba47e94ded2ec6f856daec02280fe84b
+size 36938
diff --git a/splits/unverified_first_frame.parquet b/splits/unverified_first_frame.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..efbeccd5b1dd2eb55213107f7fcd99161c075817
--- /dev/null
+++ b/splits/unverified_first_frame.parquet
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a44503e6f8bcba1df153c0fd9338f84cf3c94eebc262634ba942beb1352a3423
+size 145195
diff --git a/splits/verified_first_frame.parquet b/splits/verified_first_frame.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..12dc373b92176f872c473684d0b1e5b1d75dee4f
--- /dev/null
+++ b/splits/verified_first_frame.parquet
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af9f1e695beb137bda1bb4a1a2ab6a575a70120343816a8e5dc4d8eb67e60cd2
+size 80408
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..2f4c41c371a3c6fb33fa7ad635ec103fc7e9f560
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5691329fea3f8bb82b6f47583020958284ee007e44428098162bcf3fe9f503db
+size 765820
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/mixed.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..2e3da2ab507b62e5c7e84d181df4c17c43c3a2f1
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c4ae6b043ef19cb6fd980d3dc1563cf9f7fe245165f21de1de7363b385a0932
+size 251470
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/white.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b77a79374c429a580f5b0e33454507419a2e3dc8
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cebec925c9c76e74175b71466a4f198ae5fc9e4a68724b3d1c67b0a3c4ed191
+size 536551
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..94ff568558b35135058fde431efd12690ddf57b3
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c0d203988a4315e4b88a618be718ba601cd7a49cff3170df6741dbeb19d76bb
+size 678592
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/mixed.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..839c6a6b10facadb974dfea867a11da74bb0213f
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f40b927effa886291ff99e6a5da8ed60911770498d6255bdd536f1bcb7d1a25
+size 232975
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/white.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..53e621613ce3dfc2858a085cfed62ee02b0b1f5d
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86f78d51bf3fecb8cfd05e2a590128eba7e4593c8e300d7ca2462f535c454e4f
+size 487343
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ca3bcf039981a6c87ec52a14b969537096ade170
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:191b2db8f18ee585ba539edf705fc00e62c81b1d34745fd8b1e4cc2cf00f5591
+size 608255
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/mixed.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..40825bbcac0aba21cf507049d2f62eb9d25d5f51
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c29cd26f4550f96a7eee789e17e43ebb75acbf6a12501f8342216a3aec46425
+size 203272
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/white.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..7b98481895ac4719d02a35bb02cff442a6e89124
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95e97caf084dd891f793a58d7850b27071e194168abfac3d1d11856c55b626a2
+size 441984
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..23c4e894b60e4b3ee390c45d63762fedeb04bd8c
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:05a3b4fd81aa07aa81def156e26b209c252ba8efe67f0d2b2568ae0a3c3f11d7
+size 427543
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/mixed.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f2a875fcc91a51734bd5ce54d3aa941af80ea2eb
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f30194b137d21fdae71a2772179bddc89b9d1fc3c117fb8b16ce726195192f51
+size 152617
diff --git a/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/white.mp4 b/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..11847d28ca097ac7e776157959d7e5e18116b361
--- /dev/null
+++ b/videos/multi_view/multi4-001-forest-staggered-follow/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43d77a0b5a7dc30b06cdaa3fe521c6a4ca9ba67d4352b746b84bcbfc0f8b7b44
+size 331928
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..0c12d5a8ef6abbfb140262301b47dd09a6a65646
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c4a2c5201bb8562cf23ab0d4d9d10cf8864f5278f1858f13e8798afe568cb2f
+size 1155812
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/mixed.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9a9953612c62c4d9e97ae7831aa4510e5a441284
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e13a4eafc133c26698fa8a4a90b89bdf519d69f2a2926833c96ecf9a3892e300
+size 396502
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/white.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..fd5b5eeeeca85bb7f2ccb63a29f16920f717daf6
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af214e080ca5126c3ec9c6775d5a4e75b4e626f1db801277043c88bfd64720ae
+size 684408
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..2972eacb72e1480d4437f36c3f38fb98471a031d
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c9f295130ce15d461510c46356592668243467d7127cb6b46dfaeff2b6f1fb8
+size 1316569
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/mixed.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ed55d5b4b09efd6851d9c1b9a34f1c310327084f
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47b5c0bca6c288ecf6877f0fc37e4c51c41da87ad1f40654a6111fb6c89a6584
+size 388849
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/white.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..6417b3f028af7bc867c2f672ab5f7d3fe33e10cf
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:550b9ccd00265d86cf8ab4afd14375208bccd61993f26d9f376305b3317cd391
+size 743170
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..899f47f75dd4d43a3abd9f0d3ef986de66a7af4b
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffe35ae595d7bf181bc81b601ea3af604c0de2163e9efb3f72343cc658cd8e3c
+size 1195199
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/mixed.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b083f5cf65e66d3a42770afeed7846dd3741251c
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d68f4b205d4f6a05d81303909fb22a0e30c886e5167bf0513a05adb0d447d23c
+size 381332
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/white.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d4379c03d60158d372143e244fca663e4ed7f1b7
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:895966f951b6e769d6cb4d4567b093767299f53f24b506ca36e3d26f4bbba57b
+size 561505
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..7385b70a85318a25e6c25980f8930482fecdc0a6
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d550769c76d6cd3ce01d2a6ce301522da630c03bf6af850f77a76cb4630711e
+size 1392385
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/mixed.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d4c1955209f71b3d569251e58a05ad9c50dae0dd
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a53758186e23c249fc197940da8ac9dff2fe4e95ddae162ab12e681a856f952
+size 445222
diff --git a/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/white.mp4 b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..c0c0365d43fbc39b43e292e5a98b92016d176860
--- /dev/null
+++ b/videos/multi_view/multi4-002-jungle-opposing-pairs/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:132c77c6fe33c43c7330155a86817c4d9ff6bbf7ab012f001d2fd182515e683c
+size 685427
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..e83cea02473b22a927008b69117d2020969ef9b0
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43e2bf07aa869ba8f75f4fa20fc8e0c46f16625f66165b718d62a45364e8e35e
+size 1711658
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-a/mixed.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..a61311103bff1488e6c31d0a9071e080368c48e7
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ee3767371a77f2ec77c5ff3464608bafc20877032ea08afb2c241e935282043
+size 1554681
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-a/white.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..10f8e8eeddbb7315dc22dab49e0ece659526890b
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3ff4429ce9f10dd76ad219af1f4ac0f3f000cc58d7582efa032fa76a7a7663b
+size 1945846
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..c7f8eeb80f71dc077d98d1ba91756d737d95e0c1
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:264400c49f726a3bc610d62fa7dfb7a6d66dec042a4d7d771c7729b231596ff5
+size 1499105
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-b/mixed.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..98b5e756e1b14a19149cb87e10778f48eb4ab8d5
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d38b67794e7512a0568524eba368fb072f2c2321bd2e55a1a32caf487122deb8
+size 1328554
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-b/white.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d89287d34872c31c13879c395c4b26060622bdbb
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0286142c7c83b018d1ba5c3783408806471534016e093f7d8c68465111d69ed8
+size 1613515
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b6e9459293c7e4da6b8acdb9b81e5ae60a8994b8
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96f74583e378d9e5812c3acc5bc77fe18b26cc098f239b0a40fcbca982656e87
+size 829093
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-c/mixed.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..8653ab68667f192e5c7182d4a222db18c58f5b7e
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8f0a070f170e43e307b3fb2e7ccf648f9bc32e2ff48453616b63215d7c9a4de
+size 216637
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-c/white.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..3de21c4566df742fb37688b6e5e1f6e52e9bdc82
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af456797b595476cbbe6c062f1b09f960a6f25c8dcf91da910171cd85b60c239
+size 412201
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d40f5b745bc6c6ded82abff318f52a2eda39098e
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80db3cbb23272754730a12e907fe5192ed6923ed1588a1b6d9a52d464a14b740
+size 883318
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-d/mixed.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..935ebfa905de66f059d6f38a4717f3a3a039d1cf
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:068f7b9b04229ffabd9954bae7e038e170f992b6903a9ed059f7e2cea332fd20
+size 219387
diff --git a/videos/multi_view/multi4-003-grassland-crossroads/follow-d/white.mp4 b/videos/multi_view/multi4-003-grassland-crossroads/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d7137239e6e97c8b5afa9e101ef77aedc58d520c
--- /dev/null
+++ b/videos/multi_view/multi4-003-grassland-crossroads/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:feb30cea3afdb4f92f89c49ea59ebbb9e6a77ddfd7b6fbdd7993aa4cd9c11b8e
+size 420886
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..4d97c96f610b37539a143add0496ef18e70f222f
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5ea40a562cd9582fbcf83af47637dc4a877fdff17afe45d4fbfed324923ec37
+size 921510
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/mixed.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..30ee9fb4a1d17c34006c872b1b0b8674b6267ddf
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da69d8f2f06ca18e46651cafec9f287e54857b36f5502acae68c42106d6ed557
+size 509788
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/white.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..620250328801987819e940e8c10c9d0f06cbc43e
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e0bdb392e3b7e3cc80c43e271e688845107b26a2c0e4d2d493d1e2cb5137cd8
+size 359213
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..55c3a3b3873fdc64eb8591322f261f71b306650b
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52703dee173e6cd913482537349a67567fbbef100e28c9fdef8dc9bca1ae2694
+size 774050
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/mixed.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..02cc09a2c0ed2bf6732859e0a0d6f343b25339cd
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:436f3432e136c737055fffe85ff79b73a4c0e28714c5c3caa8625a64c9da489f
+size 560677
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/white.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9edc1c5e4c4f396861be1c70654bd6096e14396b
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b9bbcecf58d7a6d84882d916d08f4d946d68d18a5b22e047cf00bdacd5afed7
+size 407323
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..67d8cd25f30c2b6e09be260448702a7d31679271
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:758947870be8ffed061a2f06282c0fc2c68e0ef95789162033dfad69980c4b21
+size 1321505
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/mixed.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..4c0835452b8df9e2a3b43cc4257ffc1c5a60ba6f
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffd6a7ab07d8ff0121cd1eb638299d832e153e2e68e72ec76f1ee03f5f86a237
+size 624354
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/white.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..0d4c14bf82e8686c1ce11b0fc0e704064b46cc2d
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7dbc2905f06e4208a6d467be6c502f7cee2130fc2317d49ddbc3d4625ba602a
+size 1393881
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..857971c097c4b2cea6c832e59b4d464beef67656
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5d56227e18cc22d3da0b285f4c35b3b57a9fe12cfbc65129699faa33801d01f
+size 1754890
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/mixed.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f59b413a463b9297f51193bbdc3ecfa102c991a1
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cdc22c5c16a25aa781967a26828d8fdb1558a18296e025c355a5bd0d045e334
+size 767048
diff --git a/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/white.mp4 b/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..dad33413663cc291272e522bdedf7393413f0bb8
--- /dev/null
+++ b/videos/multi_view/multi4-010-underwater-converge-turn/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e057e88a2395e9350dcd325a0cd7a4dd271d5fbf0134d250a40d1fe4d56ab03
+size 1582662
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..322d11c5cc53931dccadbae7060ebbb7de3529fc
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b4be1c3f01cfd2890e46040416aa700ff5c19d7a7bc912eacac66d1e6468b2e
+size 559107
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/mixed.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..a03f30e692a79d90e9ce6072d33d0a348a1e7a36
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:237da29c0b535e9451dce612c1ab4c8ecce20ab22855a36d309bc9dc842c8db6
+size 174825
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/white.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..0787bd385a174c2450b597683ea6d121f839be6f
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aac5a3893ce1f46c84c8a363d99972493520dce5457b5fd1551579ffdf4a38f0
+size 352840
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..c12963f43b55235a24a2740c23071d0c7769bb18
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03f92ded0f2e55b722276137721bc5c3476f407c54e1560654685fbd83973b58
+size 470080
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/mixed.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ddb050d8a8397337c01760ba5e05fc95fca7b813
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:009a733785f2c029d7e3dac0ec88aea542b007bf31dde7505adce89de829bcf2
+size 164070
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/white.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..007bd4cba65d9e0fc794ff5ec6d9efd08e8c03e3
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e61d90196f37bbbf74a5b24b7558866d835a72d9adffec9f1eee874e442ac073
+size 331733
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b6749fc120cc19ba8731252e2652e3544b1b726d
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3c1ade1d40d8bcef5e2ceba60c12a9d762ff192e64ba174707b58ee71c2581c
+size 458591
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/mixed.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..461515df00d159b23203f1eb3597218f86e707de
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f5801640e97e65dbc771db4945f13b7474fbefc8a190bdf8c073a712744663c
+size 153374
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/white.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..dfd4356b3436309e27d1de3ad2f85494375fe983
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5452008fe6f02fcdad55d796626510520be02beaa5184d95fa0d7a3998ef1099
+size 329689
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b8c5d92d5c10eb0ee866abdba2b2f6acb6bb619f
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef55c529bb5cffdcd765c171a43ac67a8d905a8720c8e4dd777f04e24b6dac4f
+size 295974
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/mixed.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..7adc177b103c870725285e1441647bdd314e293e
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d73c5b384dda43b761c1bd47cd0ab48d5dd82059ce92e5468cead2be31f106c
+size 115127
diff --git a/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/white.mp4 b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f47b6f38397604359aa3c23a9ac9140ab47416a0
--- /dev/null
+++ b/videos/multi_view/multi4-013-old-town-staggered-follow/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf66b9fa4ce17fda5017915440c55090033d273c0a02ca4480adee69a96fdfc0
+size 244512
diff --git a/videos/multi_view/multi4-017-home-overtake/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-017-home-overtake/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..aa12e77d2411f7e580599e30cd462ca9410c06d6
--- /dev/null
+++ b/videos/multi_view/multi4-017-home-overtake/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:459707fb1f987c0ec500770f921ee743f12eabc1cdbbbb699760e03f4d767bd1
+size 782543
diff --git a/videos/multi_view/multi4-017-home-overtake/follow-b/mixed.mp4 b/videos/multi_view/multi4-017-home-overtake/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b617239021237227f3fbabd39eb065cf2f4d127f
--- /dev/null
+++ b/videos/multi_view/multi4-017-home-overtake/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21322432dcb3d303f2ac706695efb8f6024d7e552b7fac42f0602a69b7f53d26
+size 136169
diff --git a/videos/multi_view/multi4-017-home-overtake/follow-b/white.mp4 b/videos/multi_view/multi4-017-home-overtake/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..de4a3bc864051af18b971f6287a7446e5dba328f
--- /dev/null
+++ b/videos/multi_view/multi4-017-home-overtake/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1ceff2bb169d0e3c9e8753f99a3e5d054812db559be4b137ca2611b99e97f18
+size 391783
diff --git a/videos/multi_view/multi4-017-home-overtake/follow-c/mixed.mp4 b/videos/multi_view/multi4-017-home-overtake/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9244863c7b3e9456727179c33900451d7f490720
--- /dev/null
+++ b/videos/multi_view/multi4-017-home-overtake/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b3aeb5c52e0c3ac611f81e9179de73cde5b6c04270135318f05fea5f5883a4b
+size 214374
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..5a50db98585581be4d3223172e8c584c47ad6f85
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:932577232da4173fc86c174a9fe63c3c6316609c1db32c8c0789be89231dc5df
+size 1714890
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-a/mixed.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..08c7e5dc64fda4425ef003115d00cb0a64654e12
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a02c49208db494bd2042c4104e21468f95d203396860c16851e69c48020d331
+size 1263224
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-a/white.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..65bca60e5351750b1254f7beaefb29570130b3bb
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:614d0f50c33d0d673db0ffd47603dac108a91b062acd22c1bee20bdf66080945
+size 1562198
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ddc508c5dc941c217e4f508041bc55e150d8b348
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54f172aa1e3a896ee60cfdd3fa494e47800b76c09ee14deaa5370c33af55a59a
+size 1577690
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-b/mixed.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..611857ecb4706927e2e38ebb3dc1e0a4d4821bf5
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a53a7a94bbd4520e3d716bd41a655fbfce55f8bdf25d43e62131070e2a38111c
+size 1121578
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-b/white.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ee07e83cf805c36b6297ea16bdf5c442e5f99e35
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f48bf7d6429c0ceaa897c6053a4e9367e9c7ccf180fd30776be8bb9ecba48de4
+size 1327267
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..a710c1c0e17154b3ba743882f378f25bc29c875f
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b53eb0c855526413f0c9e9b572e30dbf1823f3d43afbce44ebf462585a69094
+size 967324
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-c/mixed.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..91a82197821267a27c60db540e6e3662697f0d9e
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24687ed4ad1551147467c955eb1f2e65a5dd0b976d94dd659facc18d7b39b413
+size 636313
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-c/white.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ba372da6a5d65e2bc3c2243bb0bdb7e3dd18fd71
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b8412402511160cdb7d7ede934ed581f131474dd2510a53de004dc108a9495f
+size 932313
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f73166b97aff84460e08a1d1d42a120be439f043
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b718b18d6688a2e3f3a419c9f8de094346866959d8428d97bad6f8522ff49bff
+size 940341
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-d/mixed.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..fcdda99cafe72a0e2e75df3360eb756b50ddad78
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f7d029adad9e9ff39a48aa87a4ba497dc1435cf06065f861a8d860299849907
+size 638560
diff --git a/videos/multi_view/multi4-021-transit-crossroads/follow-d/white.mp4 b/videos/multi_view/multi4-021-transit-crossroads/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..fe634c76a5eae754ed8155e741d15426479be352
--- /dev/null
+++ b/videos/multi_view/multi4-021-transit-crossroads/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d36cb3be0ecf12122f09f1f6a11617ad7c3757176bf0b4b4b07e91870d7b09a0
+size 880137
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-a/colored-obb.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-a/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..43883ccd94fbe22166717000a6e1380ea8655190
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-a/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f2066bd63346285a6ed2a20aab3fabe5b626436f76f7a8c1233226886a9ba7a
+size 658549
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-a/mixed.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-a/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..3159486bda9470364ac4fb5d955fbf21dc95890d
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-a/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58d266fd520f74709ea6bbc77e00d4b852681dafbcb58bb10cb8d42607e93564
+size 189381
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-a/white.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-a/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..65c450a214ad9bad633a869b14d88073daf95345
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-a/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b64986f4a13792939d8e29fdccd3a2fff72d5cf32ba5680488b4dacbb9545d57
+size 413381
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-b/colored-obb.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-b/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..dd7757496e8966e8c33b78461dd293c05efeca9d
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-b/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:907ca6043228d6b1bf9fab81b2ef7d1ea17ac67e12942cf47d8d9fa6aae1168e
+size 784301
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-b/mixed.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-b/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..68fc8b1fa7657cd397746b8f2af5e5ed7602237f
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-b/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24f07a6ccdbcac0bad37b6f92f3dbd1731680b8e465d84cb0fe0c6b06ecca4ac
+size 179037
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-b/white.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-b/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..830d7b90fac0e01f89d3869a39314fc3ffdadb8e
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-b/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c9b0e946ff6a561b2c71d0dae73587fdc1839202bbb14d25abcff4be8575d18
+size 465888
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-c/colored-obb.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-c/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..68ee2a2df3fbda1aa44ee21311ca2c3fcd6d67b7
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-c/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c47b9fb45a2fb59b5edd2f9b3ca4f86119fba1e40704097a00f280c118a0d6d
+size 963771
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-c/mixed.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-c/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..31adfbfeb58abee0e8fc434eccac2e25c7bb0ad2
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-c/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9333b1e52bf6db931ff789781521856ff8b9c81c4757628033c3dce51e3fd158
+size 219565
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-c/white.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-c/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d908cbf761312dfca20e61cbbc35b93e8347a618
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-c/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0311a1222c90f396a9c6fe2134e3d4c910ce30e4f13b1dc758a2354d507d62c9
+size 527661
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-d/colored-obb.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-d/colored-obb.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9c11d5a88888274c241a9dadbe032dfc75dc5fa4
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-d/colored-obb.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2204fa1343aee017adae403926fc62d710ea643c6ffc8ce156ff7b9036ffe276
+size 822353
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-d/mixed.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-d/mixed.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..872b8d1b569dbefe7bc925bcadff97a4ca15a9fa
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-d/mixed.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1abff9361461e72a331c44eb0b5d87b214ea6e09efd12477470ab6d7936ea5fb
+size 158365
diff --git a/videos/multi_view/multi4-023-cave-overtake/follow-d/white.mp4 b/videos/multi_view/multi4-023-cave-overtake/follow-d/white.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..10a72fd36fb2cd085fbf47d36cfa5e8e387bca37
--- /dev/null
+++ b/videos/multi_view/multi4-023-cave-overtake/follow-d/white.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ba43cc027271ef742d5e7b3a89c09d8b786b062e557d3a95ae6e96cce239b32
+size 455141