show TE-only VRAM and showcase grids

README.md

@@ -49,6 +49,34 @@ All nodes after the loader are official ComfyUI MiniMax-H3 nodes.
 
 The loader applies ARA and the recovered 8B 4096-to-5120 adapter internally. No adapter, ARA, cache, or barrier node is needed.
 
+## Showcase
+
+Matched 10-second outputs, generated with the same prompt, seed, resolution, frame count, sampler, scheduler, and step count:
+
+| Official Qwen3-VL-32B NVFP4/AWQ | Recovered Qwen3-VL-8B BF16 | Pruned-24 INT8 ConvRot |
+|---|---|---|
+| [Video](showcase/videos/official_qwen3vl_32b_nvfp4_awq_10s.mp4) | [Video](showcase/videos/recovered_qwen3vl_8b_bf16_ara_adapter_10s.mp4) | [Video](showcase/videos/pruned24_balanced_ara_int8_convrot_10s.mp4) |
+
+[30-second sequential comparison: official → recovered 8B → pruned-24](showcase/videos/minimax_h3_text_encoder_comparison_reference_recovered8b_pruned24_30s.mp4)
+
+### Twenty-frame grids
+
+The original 6720-pixel-wide combined contact sheet is preserved at [three_text_encoders_20_shots_each.png](showcase/images/three_text_encoders_20_shots_each.png). The same 60 frames are split below into three readable 5×4 grids.
+
+**Official Qwen3-VL-32B NVFP4/AWQ**
+
+![Official Qwen3-VL-32B NVFP4/AWQ — 20 sampled frames](showcase/images/official_nvfp4_20_shots_5x4.png)
+
+**Recovered Qwen3-VL-8B BF16 + ARA + adapter**
+
+![Recovered Qwen3-VL-8B BF16 — 20 sampled frames](showcase/images/recovered_8b_20_shots_5x4.png)
+
+**Pruned-24 INT8 ConvRot + balanced ARA**
+
+![Pruned-24 INT8 ConvRot — 20 sampled frames](showcase/images/pruned24_balanced_ara_20_shots_5x4.png)
+
+The showcase predates the pruned-24 NVFP4/AWQ release and does not provide a visual comparison for that storage variant. Numerical INT8-versus-NVFP4 conditioning parity is reported below.
+
 ## Required official files
 
 Download these from [Comfy-Org/MiniMax-H3](https://huggingface.co/Comfy-Org/MiniMax-H3):
@@ -59,7 +87,7 @@ Download these from [Comfy-Org/MiniMax-H3](https://huggingface.co/Comfy-Org/Mini
 
 ## Encoder variants
 
-| Node choice | Base | Extra weights | Output width | Incremental peak VRAM |
+| Node choice | Base | Extra weights | Output width | TE peak VRAM |
 |---|---|---|---:|---:|
 | `recovered_8b: recovered_8b_bf16` | Qwen3-VL-8B BF16, 24 language layers | ARA + adapter | 5120 | `10,817 MiB` (`10.56 GiB`) |
 | `recovered_8b: recovered_8b_int8_convrot` | Same recovered 8B architecture, INT8 ConvRot | ARA + adapter | 5120 | `6,333 MiB` (`6.18 GiB`) |
@@ -74,15 +102,15 @@ The old pruned-24 INT8 ConvRot artifact was physically pruned from the full 50-l
 
 The NVFP4/AWQ variant is instead a sparse copy of the pinned official `Comfy-Org/MiniMax-H3` checkpoint at revision `0543966fbdce5ba05709a8f2031c94bdba629b4a`. It preserves all retained quantized tensors exactly and removes all tensors for original language blocks `9–34`.
 
-Clean external VRAM sampling used one fresh process per encoder, the same seven-token prompt, and 100 ms `nvidia-smi` sampling. The values below are baseline-subtracted process-wide peaks:
+Clean external VRAM sampling used one fresh process per encoder, the same seven-token prompt, and 100 ms `nvidia-smi` sampling. Only the measured text-encoder contribution is reported:
 
-| Encoder | Baseline | Physical peak | Incremental peak |
-|---|---:|---:|---:|
-| Recovered 8B INT8 ConvRot | `808 MiB` | `7,141 MiB` | **`6,333 MiB` (`6.18 GiB`)** |
-| Pruned-24 NVFP4/AWQ | `503 MiB` | `10,653 MiB` | **`10,150 MiB` (`9.91 GiB`)** |
-| Recovered 8B BF16 | `808 MiB` | `11,625 MiB` | **`10,817 MiB` (`10.56 GiB`)** |
-| Pruned-24 INT8 ConvRot | `503 MiB` | `16,028 MiB` | **`15,525 MiB` (`15.16 GiB`)** |
-| Official Qwen3-VL-32B NVFP4/AWQ reference | `502 MiB` | `16,932 MiB` | **`16,430 MiB` (`16.04 GiB`)** |
+| Encoder | TE peak VRAM |
+|---|---:|
+| Recovered 8B INT8 ConvRot | **`6,333 MiB` (`6.18 GiB`)** |
+| Pruned-24 NVFP4/AWQ | **`10,150 MiB` (`9.91 GiB`)** |
+| Recovered 8B BF16 | **`10,817 MiB` (`10.56 GiB`)** |
+| Pruned-24 INT8 ConvRot | **`15,525 MiB` (`15.16 GiB`)** |
+| Official Qwen3-VL-32B NVFP4/AWQ reference | **`16,430 MiB` (`16.04 GiB`)** |
 
 The official reference is not a custom-node package choice; load it with the standard ComfyUI `CLIPLoader`. All four other rows are choices in `Load MiniMax H3 Text Encoder`. These measurements are short-prompt technical results, not general maximum requirements. Details: `evidence/released_variant_metrics.json`.
 

SHA256SUMS

@@ -1,18 +1,22 @@
 9841c64dada1d47f0ff8260af284c5db33808c59d4b0fe3c0e3a1a46fa04a4a0  .gitattributes
 c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4  LICENSE
 d71a9933d42aed06eb4dd813250c2ff13d424510339b19759723fede95a5e93b  NOTICE
-421f9ec42d715f03883f9293a4e758b553c2e139bf9029b7aaba74c25ce8b68a  README.md
+90cfd989c7e835ea44ede133c6203b84814dc6e34fef1c3f2f0e261a48bfa37c  README.md
 4c24aedf5abe9dbbb384aca8e1f9a8366b40080ee9e0961d425d05827ea8d940  evidence/comparison_validation.json
 ebf35ad3b14d354f868dfed7e7fe28a2e4bd5c26f4f14c62b9b3e44f46bb203f  evidence/cut_diagnostics.json
 649b6a8cacf1f254279b6c9087465f568554a708f0ff2e33cc23a0dab26a8bc9  evidence/final_manifest.json
 f7f7bc9b683ac65aa2eb13a270e55a8a584f3f4985b7e9ad649c3b002bb10b25  evidence/final_private_release_smoke.json
-d3c4f9053fa68e527270832262741d5dc31aac34d2ff17c251ebef22b6c0497c  evidence/pruned24_nvfp4_validation.json
+95dcdab414555dff6249b77e163e58e0e05ec5d9823118ff835b909ca967a151  evidence/pruned24_nvfp4_validation.json
 fcda410f971b45a674b023802069046b9384aa63c40f0412ec89616845141ffb  evidence/recovered_8b_int8_convrot_build_manifest.json
 4aa3d707baccd850d141ef23f568b7ef2dcc9dd8665ae3efa75c0e3b9882124c  evidence/release_loader_smoke.json
-2fc2b22b85a35d89997cc8077ebb2976c2388a2cd0b2758e506d6983c1c15807  evidence/released_variant_metrics.json
+ff2b51c89cd084525b2e563950997f75fc51a2e9c3db7022d115afbac65e2286  evidence/released_variant_metrics.json
+3b6dbeedfa60c3a63f64afc50399edcd940e87c4621f56465b6bdd2f3caf2c24  evidence/showcase_manifest.json
 7625f40f4e11dfda1f7f01e532adb65a043616b55fe21a904c57f90e54fc0dd9  evidence/technical_validation.json
 63dc23f2939827bc6c601a8fbbfad8e64f9220554b20f349748d736cdb7e17c6  evidence/three_encoder_teacher_cosine.json
-3fd7d33ed4677fed5ffbe5a8a0823405767720e549f0d68060a047330c1d7fe2  model_index.json
+b6958434654401a90c6e8150581b450141bdd0337f5f8bbd26215675c1dfb177  model_index.json
+1af5de4aa8667c61b7d63bc8e5b75d1ce7a7c9eaef48db60f76a2ea3254df779  showcase/images/official_nvfp4_20_shots_5x4.png
+f105f29b8226ac2339789d42584bf603746516d1754686cf8170c50cb886ce61  showcase/images/pruned24_balanced_ara_20_shots_5x4.png
+20980e16e72ed89ef977ded7b728982834c6fe1caaa61276d86c843691b04148  showcase/images/recovered_8b_20_shots_5x4.png
 01d731b4280c7d35b70057f2294ad4df6051ecb1215db3407fe7eca2cecc07e6  showcase/images/three_text_encoders_20_shots_each.png
 9acc6f48853f7def4a13bc2b537486a250c0ada0641427f2ca3afab80084ecd6  showcase/videos/minimax_h3_text_encoder_comparison_reference_recovered8b_pruned24_30s.mp4
 e42558dd5a7bf5b28d385c3851a4da13169881b1ea646a8c9161882c25c8053a  showcase/videos/official_qwen3vl_32b_nvfp4_awq_10s.mp4

evidence/pruned24_nvfp4_validation.json

@@ -1,26 +1,20 @@
 {
   "format": "minimax_h3_te_clean_external_vram_v1",
-  "method": "Fresh child process per variant; nvidia-smi sampled every 100 ms; identical seven-token prompt; baseline subtracted.",
+  "method": "Text-encoder peak VRAM from a fresh child process per variant; nvidia-smi sampled every 100 ms; identical seven-token prompt. Unrelated pre-existing GPU use was excluded and is not recorded.",
   "variants": [
     {
       "id": "official_nvfp4_awq",
-      "baseline_mib": 502,
-      "peak_mib": 16932,
-      "incremental_peak_mib": 16430,
+      "te_peak_vram_mib": 16430,
       "exit_code": 0
     },
     {
       "id": "pruned24_int8_convrot",
-      "baseline_mib": 503,
-      "peak_mib": 16028,
-      "incremental_peak_mib": 15525,
+      "te_peak_vram_mib": 15525,
       "exit_code": 0
     },
     {
       "id": "pruned24_nvfp4_awq",
-      "baseline_mib": 503,
-      "peak_mib": 10653,
-      "incremental_peak_mib": 10150,
+      "te_peak_vram_mib": 10150,
       "exit_code": 0
     }
   ],

evidence/released_variant_metrics.json

@@ -1,13 +1,13 @@
 {
   "format": "minimax_h3_released_variant_metrics_v1",
   "vram_method": {
-    "definition": "Fresh child process per variant; nvidia-smi sampled every 100 ms; identical seven-token prompt; baseline subtracted.",
+    "definition": "Text-encoder peak VRAM from a fresh child process per variant; nvidia-smi sampled every 100 ms; identical seven-token prompt. Unrelated pre-existing GPU use was excluded and is not recorded.",
     "unit": "MiB",
-    "scope": "External process-wide peak GPU memory during text-conditioning smoke.",
+    "scope": "Peak GPU memory attributable to the text-encoder conditioning smoke.",
     "hardware": "NVIDIA GeForce RTX 5090 Laptop GPU",
     "limitations": [
       "Short text-conditioning smoke; longer or multimodal conditioning can require more memory.",
-      "Process-wide GPU memory is not a tensor-residency attribution."
+      "TE peak VRAM is an end-to-end runtime measurement, not a tensor-residency attribution."
     ]
   },
   "cosine_definitions": {
@@ -17,20 +17,16 @@
   "released_node_variants": [
     {
       "node_choice": "recovered_8b: recovered_8b_bf16",
-      "baseline_mib": 808,
-      "physical_peak_mib": 11625,
-      "incremental_peak_mib": 10817,
-      "incremental_peak_gib": 10.56,
+      "te_peak_vram_mib": 10817,
+      "te_peak_vram_gib": 10.56,
       "teacher_reference_cosine": 0.938678461010568,
       "quantization_parity_reference": "recovered_8b: recovered_8b_int8_convrot",
       "quantization_parity_cosine": 0.9999650716781616
     },
     {
       "node_choice": "recovered_8b: recovered_8b_int8_convrot",
-      "baseline_mib": 808,
-      "physical_peak_mib": 7141,
-      "incremental_peak_mib": 6333,
-      "incremental_peak_gib": 6.18,
+      "te_peak_vram_mib": 6333,
+      "te_peak_vram_gib": 6.18,
       "teacher_reference_cosine": null,
       "teacher_reference_note": "Not directly measured in the original matched teacher-reference comparison; do not substitute quantization parity.",
       "quantization_parity_reference": "recovered_8b: recovered_8b_bf16",
@@ -38,20 +34,16 @@
     },
     {
       "node_choice": "pruned24_balanced: int8_convrot",
-      "baseline_mib": 503,
-      "physical_peak_mib": 16028,
-      "incremental_peak_mib": 15525,
-      "incremental_peak_gib": 15.16,
+      "te_peak_vram_mib": 15525,
+      "te_peak_vram_gib": 15.16,
       "teacher_reference_cosine": 0.9404705529707683,
       "quantization_parity_reference": "pruned24_balanced: nvfp4_awq",
       "quantization_parity_cosine": 0.9999949932098389
     },
     {
       "node_choice": "pruned24_balanced: nvfp4_awq",
-      "baseline_mib": 503,
-      "physical_peak_mib": 10653,
-      "incremental_peak_mib": 10150,
-      "incremental_peak_gib": 9.91,
+      "te_peak_vram_mib": 10150,
+      "te_peak_vram_gib": 9.91,
       "teacher_reference_cosine": null,
       "teacher_reference_note": "Not directly measured in the original matched teacher-reference comparison; do not substitute quantization parity.",
       "quantization_parity_reference": "pruned24_balanced: int8_convrot",
@@ -60,10 +52,8 @@
   ],
   "official_reference": {
     "id": "official_qwen3vl_32b_nvfp4_awq",
-    "baseline_mib": 502,
-    "physical_peak_mib": 16932,
-    "incremental_peak_mib": 16430,
-    "incremental_peak_gib": 16.04,
+    "te_peak_vram_mib": 16430,
+    "te_peak_vram_gib": 16.04,
     "teacher_reference_cosine": 1.0,
     "note": "Comparison reference; loaded with the standard ComfyUI CLIPLoader, not a custom-node package choice."
   },

evidence/showcase_manifest.json

@@ -0,0 +1,155 @@
+{
+  "format": "minimax_h3_showcase_manifest_v1",
+  "matched_encoder_order": [
+    "official_nvfp4_awq",
+    "recovered_8b_bf16_ara_adapter",
+    "pruned24_balanced_ara_int8_convrot"
+  ],
+  "assets": [
+    {
+      "path": "showcase/images/official_nvfp4_20_shots_5x4.png",
+      "bytes": 1730101,
+      "sha256": "1af5de4aa8667c61b7d63bc8e5b75d1ce7a7c9eaef48db60f76a2ea3254df779",
+      "type": "image",
+      "width": 1600,
+      "height": 836
+    },
+    {
+      "path": "showcase/images/pruned24_balanced_ara_20_shots_5x4.png",
+      "bytes": 1760765,
+      "sha256": "f105f29b8226ac2339789d42584bf603746516d1754686cf8170c50cb886ce61",
+      "type": "image",
+      "width": 1600,
+      "height": 836
+    },
+    {
+      "path": "showcase/images/recovered_8b_20_shots_5x4.png",
+      "bytes": 1743298,
+      "sha256": "20980e16e72ed89ef977ded7b728982834c6fe1caaa61276d86c843691b04148",
+      "type": "image",
+      "width": 1600,
+      "height": 836
+    },
+    {
+      "path": "showcase/images/three_text_encoders_20_shots_each.png",
+      "bytes": 5270610,
+      "sha256": "01d731b4280c7d35b70057f2294ad4df6051ecb1215db3407fe7eca2cecc07e6",
+      "type": "image",
+      "width": 6720,
+      "height": 660
+    },
+    {
+      "path": "showcase/videos/minimax_h3_text_encoder_comparison_reference_recovered8b_pruned24_30s.mp4",
+      "bytes": 11602057,
+      "sha256": "9acc6f48853f7def4a13bc2b537486a250c0ada0641427f2ca3afab80084ecd6",
+      "type": "video",
+      "probe": {
+        "programs": [],
+        "streams": [
+          {
+            "codec_name": "h264",
+            "codec_type": "video",
+            "width": 1344,
+            "height": 768,
+            "r_frame_rate": "24/1"
+          },
+          {
+            "codec_name": "aac",
+            "codec_type": "audio",
+            "r_frame_rate": "0/0"
+          }
+        ],
+        "format": {
+          "duration": "30.032000",
+          "size": "11602057"
+        }
+      }
+    },
+    {
+      "path": "showcase/videos/official_qwen3vl_32b_nvfp4_awq_10s.mp4",
+      "bytes": 3967207,
+      "sha256": "e42558dd5a7bf5b28d385c3851a4da13169881b1ea646a8c9161882c25c8053a",
+      "type": "video",
+      "probe": {
+        "programs": [],
+        "streams": [
+          {
+            "codec_name": "h264",
+            "codec_type": "video",
+            "width": 1344,
+            "height": 768,
+            "r_frame_rate": "24/1"
+          },
+          {
+            "codec_name": "aac",
+            "codec_type": "audio",
+            "r_frame_rate": "0/0"
+          }
+        ],
+        "format": {
+          "duration": "10.000000",
+          "size": "3967207"
+        }
+      }
+    },
+    {
+      "path": "showcase/videos/pruned24_balanced_ara_int8_convrot_10s.mp4",
+      "bytes": 4212574,
+      "sha256": "8091a8ab44519e7629d5e36c26940764f98d3abf0a0b2c7937f81a940a094124",
+      "type": "video",
+      "probe": {
+        "programs": [],
+        "streams": [
+          {
+            "codec_name": "h264",
+            "codec_type": "video",
+            "width": 1344,
+            "height": 768,
+            "r_frame_rate": "24/1"
+          },
+          {
+            "codec_name": "aac",
+            "codec_type": "audio",
+            "r_frame_rate": "0/0"
+          }
+        ],
+        "format": {
+          "duration": "10.000000",
+          "size": "4212574"
+        }
+      }
+    },
+    {
+      "path": "showcase/videos/recovered_qwen3vl_8b_bf16_ara_adapter_10s.mp4",
+      "bytes": 3713830,
+      "sha256": "1ab0cbc71335418921c99a74191f50d77f5aa58f3a5d6a035266758ada34020f",
+      "type": "video",
+      "probe": {
+        "programs": [],
+        "streams": [
+          {
+            "codec_name": "h264",
+            "codec_type": "video",
+            "width": 1344,
+            "height": 768,
+            "r_frame_rate": "24/1"
+          },
+          {
+            "codec_name": "aac",
+            "codec_type": "audio",
+            "r_frame_rate": "0/0"
+          }
+        ],
+        "format": {
+          "duration": "10.000000",
+          "size": "3713830"
+        }
+      }
+    }
+  ],
+  "limitations": [
+    "No visual-quality winner is claimed.",
+    "The requested hard cuts at 3 and 7 seconds did not pass for any encoder.",
+    "The pruned-24 NVFP4/AWQ variant was released after this visual showcase; its INT8-versus-NVFP4 evidence is numerical conditioning parity only."
+  ]
+}

model_index.json

@@ -23,6 +23,7 @@
     }
   ],
   "released_variant_metrics": "evidence/released_variant_metrics.json",
+  "showcase_manifest": "evidence/showcase_manifest.json",
   "required_custom_node": "https://github.com/kgonia/ComfyUI-MiniMaxH3TextEncoders",
   "official_downstream_nodes": ["MiniMaxH3ImageToVideo", "MiniMaxH3SigmaShift", "UNETLoader", "VAELoader", "SamplerCustomAdvanced", "VAEDecode", "VAEDecodeAudio", "CreateVideo", "SaveVideo"]
 }

release_manifest.json

@@ -16,9 +16,9 @@
       "sha256": "d71a9933d42aed06eb4dd813250c2ff13d424510339b19759723fede95a5e93b"
     },
     {
-      "bytes": 7791,
+      "bytes": 9111,
       "path": "README.md",
-      "sha256": "421f9ec42d715f03883f9293a4e758b553c2e139bf9029b7aaba74c25ce8b68a"
+      "sha256": "90cfd989c7e835ea44ede133c6203b84814dc6e34fef1c3f2f0e261a48bfa37c"
     },
     {
       "bytes": 6874,
@@ -41,9 +41,9 @@
       "sha256": "f7f7bc9b683ac65aa2eb13a270e55a8a584f3f4985b7e9ad649c3b002bb10b25"
     },
     {
-      "bytes": 1723,
+      "bytes": 1629,
       "path": "evidence/pruned24_nvfp4_validation.json",
-      "sha256": "d3c4f9053fa68e527270832262741d5dc31aac34d2ff17c251ebef22b6c0497c"
+      "sha256": "95dcdab414555dff6249b77e163e58e0e05ec5d9823118ff835b909ca967a151"
     },
     {
       "bytes": 1786,
@@ -56,9 +56,14 @@
       "sha256": "4aa3d707baccd850d141ef23f568b7ef2dcc9dd8665ae3efa75c0e3b9882124c"
     },
     {
-      "bytes": 3287,
+      "bytes": 3044,
       "path": "evidence/released_variant_metrics.json",
-      "sha256": "2fc2b22b85a35d89997cc8077ebb2976c2388a2cd0b2758e506d6983c1c15807"
+      "sha256": "ff2b51c89cd084525b2e563950997f75fc51a2e9c3db7022d115afbac65e2286"
+    },
+    {
+      "bytes": 4358,
+      "path": "evidence/showcase_manifest.json",
+      "sha256": "3b6dbeedfa60c3a63f64afc50399edcd940e87c4621f56465b6bdd2f3caf2c24"
     },
     {
       "bytes": 5784,
@@ -71,9 +76,24 @@
       "sha256": "63dc23f2939827bc6c601a8fbbfad8e64f9220554b20f349748d736cdb7e17c6"
     },
     {
-      "bytes": 1444,
+      "bytes": 1502,
       "path": "model_index.json",
-      "sha256": "3fd7d33ed4677fed5ffbe5a8a0823405767720e549f0d68060a047330c1d7fe2"
+      "sha256": "b6958434654401a90c6e8150581b450141bdd0337f5f8bbd26215675c1dfb177"
+    },
+    {
+      "bytes": 1730101,
+      "path": "showcase/images/official_nvfp4_20_shots_5x4.png",
+      "sha256": "1af5de4aa8667c61b7d63bc8e5b75d1ce7a7c9eaef48db60f76a2ea3254df779"
+    },
+    {
+      "bytes": 1760765,
+      "path": "showcase/images/pruned24_balanced_ara_20_shots_5x4.png",
+      "sha256": "f105f29b8226ac2339789d42584bf603746516d1754686cf8170c50cb886ce61"
+    },
+    {
+      "bytes": 1743298,
+      "path": "showcase/images/recovered_8b_20_shots_5x4.png",
+      "sha256": "20980e16e72ed89ef977ded7b728982834c6fe1caaa61276d86c843691b04148"
     },
     {
       "bytes": 5270610,
@@ -150,7 +170,7 @@
   "scope": {
     "adapters": true,
     "diffusion_models": false,
-    "showcase_matrix_images": 1,
+    "showcase_matrix_images": 4,
     "showcase_videos": 4,
     "text_encoders": true,
     "vaes": false
@@ -173,6 +193,7 @@
     },
     "release_loader_smoke": "passed",
     "released_variant_metrics": "evidence/released_variant_metrics.json",
+    "showcase_manifest": "evidence/showcase_manifest.json",
     "variants": {
       "pruned24_balanced: int8_convrot": {
         "conditioning_dtype": "torch.float32",