MiniMax H3
comfyui
video
steering
dials
matlod commited on
Commit
4a19e96
·
verified ·
1 Parent(s): 6ac4c6d

Pika H3 Flex: motion dial sweep and de-rope videos

Browse files
.gitattributes CHANGED
@@ -36,3 +36,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
36
  assets/motion_dial_demo.mp4 filter=lfs diff=lfs merge=lfs -text
37
  assets/motion_dial_dog.gif filter=lfs diff=lfs merge=lfs -text
38
  assets/pika_dread_compare.mp4 filter=lfs diff=lfs merge=lfs -text
 
 
 
36
  assets/motion_dial_demo.mp4 filter=lfs diff=lfs merge=lfs -text
37
  assets/motion_dial_dog.gif filter=lfs diff=lfs merge=lfs -text
38
  assets/pika_dread_compare.mp4 filter=lfs diff=lfs merge=lfs -text
39
+ assets/pika_derope_compare.mp4 filter=lfs diff=lfs merge=lfs -text
40
+ assets/pika_dial_sweep.mp4 filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -63,6 +63,25 @@ seed; audio from +1). The dials work on Pika unchanged and keep the prompt and c
63
  useful range for motion and evil alike. Ready-made workflows (Pika + motion dial + evil slot, with and without de-rope):
64
  **[ComfyUI-Dials › Pika H3 Flex + Dials](https://github.com/matlowai/ComfyUI-Dials#pika-h3-flex--dials)**.
65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
66
  ## What makes a dial different
67
 
68
  - **Not a LoRA or slider LoRA:** nothing is trained. A dial is computed from prompt pairs in about a minute and only
 
63
  useful range for motion and evil alike. Ready-made workflows (Pika + motion dial + evil slot, with and without de-rope):
64
  **[ComfyUI-Dials › Pika H3 Flex + Dials](https://github.com/matlowai/ComfyUI-Dials#pika-h3-flex--dials)**.
65
 
66
+ ### Motion dial on Pika: 0 to +3
67
+
68
+ <video controls autoplay loop muted src="https://huggingface.co/MATLOWAI/H3-Dials/resolve/main/assets/pika_dial_sweep.mp4"></video>
69
+
70
+ 1280x704, same prompt and seed. **+0.5 / +1.0**: same scene, more action (+20 % / +23 % camera-removed motion).
71
+ **+1.5 / +2.0**: the scene restages and characters morph. **+3**: loses the prompt.
72
+
73
+ ### De-rope vs no de-rope
74
+
75
+ <video controls autoplay loop muted src="https://huggingface.co/MATLOWAI/H3-Dials/resolve/main/assets/pika_derope_compare.mp4"></video>
76
+
77
+ Motion +1.0. De-rope finds stretches that move too fast for the frame rate, re-renders them slowed down and maps them
78
+ back to real time, keeping the original audio. The workflow default re-enters at inject 0.48; a 6+6 schedule is
79
+ included as an experimental workflow. At 1280x704, a de-roped Pika clip takes about 5.5–6 min, about the same as one
80
+ 25-step render of the base model without de-rope (about 5.5 min).
81
+
82
+ *These two videos were rendered with an earlier wording of the example prompt, before the magical-girl rewrite.
83
+ Re-renders with the current wording are coming.*
84
+
85
  ## What makes a dial different
86
 
87
  - **Not a LoRA or slider LoRA:** nothing is trained. A dial is computed from prompt pairs in about a minute and only
assets/pika_derope_compare.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f9e68c77bc64fe3965faca627db67f6be2f47656205c6ccf3aecfb5b460fa4a
3
+ size 1985713
assets/pika_dial_sweep.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d116da3687f1794b53e35f366f8a52eb168af386074efead2dbeabd754e7b018
3
+ size 2819750