ComfyUI workflow for 8 GB GPUs (5 s); pick-by-GPU table; --low-vram on 8 GB / GPU-size presets

Adds minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.{json,mp4} and its screenshot; model card: 8 GB section, workflow-by-GPU table, inference.py --low-vram on 8 GB GPUs.

.gitattributes

@@ -38,3 +38,5 @@ minimax_h3/workflows/dmad_h3_4step_15s_podcast.mp4 filter=lfs diff=lfs merge=lfs
 minimax_h3/workflows/dmad_h3_8step_15s_wok.mp4 filter=lfs diff=lfs merge=lfs -text
 minimax_h3/workflows/screenshot_4step_15s.png filter=lfs diff=lfs merge=lfs -text
 minimax_h3/workflows/screenshot_8step_15s.png filter=lfs diff=lfs merge=lfs -text
+minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.mp4 filter=lfs diff=lfs merge=lfs -text
+minimax_h3/workflows/screenshot_4step_5s_8gb.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -62,7 +62,7 @@ token-refiner blocks (312 modules). alpha = rank = 128. The safetensors metadata
 
 The inference code lives in the [code repository](https://github.com/Yzmblog/DMAD): `inference.py` with the sampler
 the paper used (re-noise step rule) and a Diffusers-pipeline example; its README covers the environment. Sampling
-settings: 4 steps, time shift 12 (video) and 2 (audio), no classifier-free guidance, 124 frames at 24 fps. `--low-vram` runs it on a 16 GB GPU.
+settings: 4 steps, time shift 12 (video) and 2 (audio), no classifier-free guidance, 124 frames at 24 fps. `--low-vram` runs it on an 8 GB GPU (5 s; 15 s needs 12 GB; `--low-vram 8 / 12 / 16 / 24` picks the settings for your GPU, same output).
 
 ```bash
 git clone https://github.com/Yzmblog/DMAD.git && cd DMAD   # code + environment setup (see its README)
@@ -87,12 +87,22 @@ wget -P /path/to/ComfyUI/models/loras https://huggingface.co/ZhengmingYu/DMAD/re
 wget -P /path/to/ComfyUI/models/loras https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/dmad_minimax_h3_4step_lora_critic_comfyui.safetensors
 ```
 
-### Ready-to-run workflows (16 GB GPU and up)
+### Ready-to-run workflows (8 GB GPU and up)
 
-Two complete text-to-audio-video workflows in [`minimax_h3/workflows`](https://huggingface.co/ZhengmingYu/DMAD/tree/main/minimax_h3/workflows)
-that make 15 s of 1344x768 video with stereo audio on a 16 GB GPU or larger. Load the `.json`, or drag the example `.mp4` (it
-embeds the workflow) into ComfyUI; missing models are offered for download from links stored in the workflow. Both
-sample with stock nodes (`lcm` + `simple`, `full_critic` LoRA); the ComfyUI-DMAD nodes are not needed.
+Complete text-to-audio-video workflows in [`minimax_h3/workflows`](https://huggingface.co/ZhengmingYu/DMAD/tree/main/minimax_h3/workflows):
+two that make 15 s of 1344x768 video with stereo audio on a 16 GB GPU or larger, and a 5 s one for 8 GB GPUs. Load the
+`.json`, or drag the example `.mp4` (it embeds the workflow) into ComfyUI; missing models are offered for download from
+links stored in the workflow. All sample with stock nodes (`lcm` + `simple`, `full_critic` LoRA); the ComfyUI-DMAD nodes
+are not needed.
+
+| GPU | workflow | video |
+|---|---|---|
+| 24 GB | [`dmad_h3_4step_15s_podcast.json`](https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/workflows/dmad_h3_4step_15s_podcast.json) (4 steps) or [`dmad_h3_8step_15s_wok.json`](https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/workflows/dmad_h3_8step_15s_wok.json) (8 steps) | 15 s |
+| 16 GB, 22 GB | the same two workflows, unchanged (see below) | 15 s |
+| 8 GB | [`dmad_h3_4step_5s_8gb_guitar.json`](https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.json) (see below) | 5 s |
+
+The 24 GB and 16 GB rows share the workflow files: ComfyUI keeps as much of the model on the GPU as fits and streams
+the rest from system memory, so the same workflow adapts to the card (and gives the same video).
 
 | | 4 steps | 8 steps |
 |---|---|---|
@@ -126,8 +136,28 @@ with the PyTorch allocator capped at 24 or 14 GiB; a consumer GPU is slower, and
 card's desktop and CUDA context): ComfyUI's dynamic VRAM keeps less of the model resident and streams more of it from
 system memory, and the videos are bit-identical to the 24 GiB runs. About 46 GiB of system memory is in use, so 64 GB
 of RAM is recommended. Outside ComfyUI, `inference.py --low-vram` in the
-[code repository](https://github.com/Yzmblog/DMAD) makes 5 s and 15 s videos in under 14 GiB of GPU memory, bit-identical
-to its larger-GPU runs.
+[code repository](https://github.com/Yzmblog/DMAD) makes 5 s videos in under 7 GiB and 15 s videos in 12 GiB of GPU memory,
+bit-identical to its larger-GPU runs.
+
+**8 GB GPUs (5 s).** [`dmad_h3_4step_5s_8gb_guitar.json`](https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.json) is the 4-step
+workflow set up for 8 GB GPUs: 5 s instead of 15 s, the `H3 Memory Optimization` node in its lowest-memory setting
+(`Attention memory mode` Lower VRAM (slower), `Activation chunk rows` 1024), and the `H3 AIMDO Residency Limiter` node of
+the same pack (`VBAR residency budget` 0 blocks), which streams every transformer block from system memory. Models and
+sampling are the same as above.
+
+| | 4 steps, 5 s |
+|---|---|
+| Workflow | [`dmad_h3_4step_5s_8gb_guitar.json`](https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.json) |
+| Example (workflow embedded) | [`dmad_h3_4step_5s_8gb_guitar.mp4`](https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.mp4), seed 1 |
+| Output | 1344x768, 124 frames = 5 s at 24 fps, stereo audio |
+| Peak GPU memory, 7 GiB cap (8 GB GPUs) | 7.7 GiB |
+| Peak GPU memory, 5 GiB cap | 5.9 GiB |
+| Time (either cap) | about 210 s, of which sampling 152 s |
+
+The video is the same at both caps, bit for bit; the peaks include memory ComfyUI's dynamic VRAM allocates outside the
+PyTorch allocator. About 45 GiB of system memory is in use (64 GB of RAM recommended). The memory settings change the
+attention arithmetic, so the same prompt and seed do not reproduce the video the 16 GB+ settings make. 15 s does not
+fit in 8 GB.
 
 ## Wan2.1 (text-to-video)
 

minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.json

@@ -0,0 +1,1214 @@
+{
+ "id": "dmad-minimax-h3-t2av-8gb-4step",
+ "revision": 0,
+ "last_node_id": 19,
+ "last_link_id": 21,
+ "nodes": [
+  {
+   "id": 19,
+   "type": "H3AIMDOResidencyLimiter",
+   "pos": [
+    0,
+    1050
+   ],
+   "size": [
+    430,
+    60
+   ],
+   "flags": {},
+   "order": 17,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "model",
+     "type": "MODEL",
+     "link": 4
+    }
+   ],
+   "outputs": [
+    {
+     "name": "MODEL",
+     "type": "MODEL",
+     "links": [
+      5,
+      6
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "h3-optimizations",
+    "ver": "0.2.45",
+    "Node name for S&R": "H3AIMDOResidencyLimiter"
+   },
+   "widgets_values": [
+    "0 blocks"
+   ]
+  },
+  {
+   "id": 1,
+   "type": "UNETLoader",
+   "pos": [
+    0,
+    0
+   ],
+   "size": [
+    430,
+    90
+   ],
+   "flags": {},
+   "order": 0,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "unet_name",
+     "type": "COMBO",
+     "widget": {
+      "name": "unet_name"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "MODEL",
+     "type": "MODEL",
+     "links": [
+      1
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "UNETLoader",
+    "models": [
+     {
+      "name": "minimax_h3_fl2va_pruned_int8_convrot.safetensors",
+      "url": "https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors",
+      "directory": "diffusion_models"
+     }
+    ]
+   },
+   "widgets_values": [
+    "minimax_h3_fl2va_pruned_int8_convrot.safetensors",
+    "default"
+   ]
+  },
+  {
+   "id": 2,
+   "type": "LoraLoaderModelOnly",
+   "pos": [
+    0,
+    140
+   ],
+   "size": [
+    430,
+    90
+   ],
+   "flags": {},
+   "order": 1,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "model",
+     "type": "MODEL",
+     "link": 1
+    },
+    {
+     "name": "lora_name",
+     "type": "COMBO",
+     "widget": {
+      "name": "lora_name"
+     },
+     "link": null
+    },
+    {
+     "name": "strength_model",
+     "type": "FLOAT",
+     "widget": {
+      "name": "strength_model"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "MODEL",
+     "type": "MODEL",
+     "links": [
+      2
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "LoraLoaderModelOnly",
+    "models": [
+     {
+      "name": "dmad_minimax_h3_4step_full_critic_comfyui.safetensors",
+      "url": "https://huggingface.co/ZhengmingYu/DMAD/resolve/main/minimax_h3/dmad_minimax_h3_4step_full_critic_comfyui.safetensors",
+      "directory": "loras"
+     }
+    ]
+   },
+   "widgets_values": [
+    "dmad_minimax_h3_4step_full_critic_comfyui.safetensors",
+    1
+   ]
+  },
+  {
+   "id": 3,
+   "type": "MiniMaxH3SigmaShift",
+   "pos": [
+    0,
+    280
+   ],
+   "size": [
+    430,
+    90
+   ],
+   "flags": {},
+   "order": 2,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "model",
+     "type": "MODEL",
+     "link": 2
+    },
+    {
+     "name": "shift_video",
+     "type": "FLOAT",
+     "widget": {
+      "name": "shift_video"
+     },
+     "link": null
+    },
+    {
+     "name": "shift_audio",
+     "type": "FLOAT",
+     "widget": {
+      "name": "shift_audio"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "MODEL",
+     "type": "MODEL",
+     "links": [
+      3
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.35.1",
+    "Node name for S&R": "MiniMaxH3SigmaShift"
+   },
+   "widgets_values": [
+    12,
+    2
+   ]
+  },
+  {
+   "id": 4,
+   "type": "CLIPLoader",
+   "pos": [
+    0,
+    420
+   ],
+   "size": [
+    430,
+    110
+   ],
+   "flags": {},
+   "order": 3,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "clip_name",
+     "type": "COMBO",
+     "widget": {
+      "name": "clip_name"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "CLIP",
+     "type": "CLIP",
+     "links": [
+      7
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "CLIPLoader",
+    "models": [
+     {
+      "name": "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors",
+      "url": "https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors",
+      "directory": "text_encoders"
+     }
+    ]
+   },
+   "widgets_values": [
+    "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors",
+    "minimax",
+    "default"
+   ]
+  },
+  {
+   "id": 5,
+   "type": "VAELoader",
+   "pos": [
+    0,
+    580
+   ],
+   "size": [
+    430,
+    60
+   ],
+   "flags": {},
+   "order": 4,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "vae_name",
+     "type": "COMBO",
+     "widget": {
+      "name": "vae_name"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "VAE",
+     "type": "VAE",
+     "links": [
+      8,
+      16
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "VAELoader",
+    "models": [
+     {
+      "name": "minimax_h3_video_vae_fp16.safetensors",
+      "url": "https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_fp16.safetensors",
+      "directory": "vae"
+     }
+    ]
+   },
+   "widgets_values": [
+    "minimax_h3_video_vae_fp16.safetensors"
+   ]
+  },
+  {
+   "id": 6,
+   "type": "VAELoader",
+   "pos": [
+    0,
+    690
+   ],
+   "size": [
+    430,
+    60
+   ],
+   "flags": {},
+   "order": 5,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "vae_name",
+     "type": "COMBO",
+     "widget": {
+      "name": "vae_name"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "VAE",
+     "type": "VAE",
+     "links": [
+      18
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "VAELoader",
+    "models": [
+     {
+      "name": "minimax_h3_audio_vae_fp32.safetensors",
+      "url": "https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors",
+      "directory": "vae"
+     }
+    ]
+   },
+   "widgets_values": [
+    "minimax_h3_audio_vae_fp32.safetensors"
+   ]
+  },
+  {
+   "id": 7,
+   "type": "MiniMaxH3ImageToVideo",
+   "pos": [
+    480,
+    0
+   ],
+   "size": [
+    520,
+    560
+   ],
+   "flags": {},
+   "order": 6,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "clip",
+     "type": "CLIP",
+     "link": 7
+    },
+    {
+     "name": "vae",
+     "type": "VAE",
+     "link": 8
+    },
+    {
+     "name": "first_frame",
+     "shape": 7,
+     "type": "IMAGE",
+     "link": null
+    },
+    {
+     "name": "last_frame",
+     "shape": 7,
+     "type": "IMAGE",
+     "link": null
+    },
+    {
+     "name": "prompt",
+     "type": "STRING",
+     "widget": {
+      "name": "prompt"
+     },
+     "link": null
+    },
+    {
+     "name": "width",
+     "type": "INT",
+     "widget": {
+      "name": "width"
+     },
+     "link": null
+    },
+    {
+     "name": "height",
+     "type": "INT",
+     "widget": {
+      "name": "height"
+     },
+     "link": null
+    },
+    {
+     "name": "length",
+     "type": "INT",
+     "widget": {
+      "name": "length"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "positive",
+     "type": "CONDITIONING",
+     "links": [
+      9
+     ]
+    },
+    {
+     "name": "LATENT",
+     "type": "LATENT",
+     "links": [
+      14
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "MiniMaxH3ImageToVideo"
+   },
+   "widgets_values": [
+    "integrated_multimodal_description: [Shot 1] Live-action, golden-hour city rooftop with the skyline glowing behind, a medium shot frames a young man with a denim jacket sitting on a ledge and strumming a natural-wood acoustic guitar with \"DMAD\" painted in bold white letters on its body below the sound hole, the warm sun flaring behind him. [Shot 2] At 00:02.500, the camera slowly arcs to a close-up as he sings a soft line with his eyes closed, and the young man (S1) sings: <d>[English] Golden light, stay with me tonight.</d>\n\noverall_soundscape: A gentle breeze, distant city traffic far below, the bright strum of the acoustic guitar, and his soft live vocal.\n\nnon_diegetic_music: N/A",
+    1344,
+    768,
+    124
+   ]
+  },
+  {
+   "id": 8,
+   "type": "RandomNoise",
+   "pos": [
+    1050,
+    0
+   ],
+   "size": [
+    320,
+    90
+   ],
+   "flags": {},
+   "order": 7,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "noise_seed",
+     "type": "INT",
+     "widget": {
+      "name": "noise_seed"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "NOISE",
+     "type": "NOISE",
+     "links": [
+      10
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "RandomNoise"
+   },
+   "widgets_values": [
+    1,
+    "fixed"
+   ]
+  },
+  {
+   "id": 9,
+   "type": "BasicGuider",
+   "pos": [
+    1050,
+    140
+   ],
+   "size": [
+    320,
+    60
+   ],
+   "flags": {},
+   "order": 8,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "model",
+     "type": "MODEL",
+     "link": 5
+    },
+    {
+     "name": "conditioning",
+     "type": "CONDITIONING",
+     "link": 9
+    }
+   ],
+   "outputs": [
+    {
+     "name": "GUIDER",
+     "type": "GUIDER",
+     "links": [
+      11
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "BasicGuider"
+   }
+  },
+  {
+   "id": 10,
+   "type": "KSamplerSelect",
+   "pos": [
+    1050,
+    250
+   ],
+   "size": [
+    320,
+    60
+   ],
+   "flags": {},
+   "order": 9,
+   "mode": 0,
+   "inputs": [],
+   "outputs": [
+    {
+     "name": "SAMPLER",
+     "type": "SAMPLER",
+     "links": [
+      12
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "KSamplerSelect"
+   },
+   "widgets_values": [
+    "lcm"
+   ]
+  },
+  {
+   "id": 11,
+   "type": "BasicScheduler",
+   "pos": [
+    1050,
+    360
+   ],
+   "size": [
+    320,
+    110
+   ],
+   "flags": {},
+   "order": 10,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "model",
+     "type": "MODEL",
+     "link": 6
+    },
+    {
+     "name": "steps",
+     "type": "INT",
+     "widget": {
+      "name": "steps"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "SIGMAS",
+     "type": "SIGMAS",
+     "links": [
+      13
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "BasicScheduler"
+   },
+   "widgets_values": [
+    "simple",
+    4,
+    1
+   ]
+  },
+  {
+   "id": 12,
+   "type": "SamplerCustomAdvanced",
+   "pos": [
+    1050,
+    520
+   ],
+   "size": [
+    320,
+    110
+   ],
+   "flags": {},
+   "order": 11,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "noise",
+     "type": "NOISE",
+     "link": 10
+    },
+    {
+     "name": "guider",
+     "type": "GUIDER",
+     "link": 11
+    },
+    {
+     "name": "sampler",
+     "type": "SAMPLER",
+     "link": 12
+    },
+    {
+     "name": "sigmas",
+     "type": "SIGMAS",
+     "link": 13
+    },
+    {
+     "name": "latent_image",
+     "type": "LATENT",
+     "link": 14
+    }
+   ],
+   "outputs": [
+    {
+     "name": "output",
+     "type": "LATENT",
+     "links": [
+      15,
+      17
+     ]
+    },
+    {
+     "name": "denoised_output",
+     "type": "LATENT",
+     "links": []
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "SamplerCustomAdvanced"
+   }
+  },
+  {
+   "id": 13,
+   "type": "VAEDecode",
+   "pos": [
+    1420,
+    0
+   ],
+   "size": [
+    260,
+    50
+   ],
+   "flags": {
+    "collapsed": false
+   },
+   "order": 12,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "samples",
+     "type": "LATENT",
+     "link": 15
+    },
+    {
+     "name": "vae",
+     "type": "VAE",
+     "link": 16
+    }
+   ],
+   "outputs": [
+    {
+     "name": "IMAGE",
+     "type": "IMAGE",
+     "links": [
+      19
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "VAEDecode"
+   }
+  },
+  {
+   "id": 14,
+   "type": "VAEDecodeAudio",
+   "pos": [
+    1420,
+    100
+   ],
+   "size": [
+    260,
+    50
+   ],
+   "flags": {
+    "collapsed": false
+   },
+   "order": 13,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "samples",
+     "type": "LATENT",
+     "link": 17
+    },
+    {
+     "name": "vae",
+     "type": "VAE",
+     "link": 18
+    }
+   ],
+   "outputs": [
+    {
+     "name": "AUDIO",
+     "type": "AUDIO",
+     "links": [
+      20
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "VAEDecodeAudio"
+   }
+  },
+  {
+   "id": 15,
+   "type": "CreateVideo",
+   "pos": [
+    1420,
+    200
+   ],
+   "size": [
+    260,
+    130
+   ],
+   "flags": {},
+   "order": 14,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "images",
+     "type": "IMAGE",
+     "link": 19
+    },
+    {
+     "name": "audio",
+     "shape": 7,
+     "type": "AUDIO",
+     "link": 20
+    }
+   ],
+   "outputs": [
+    {
+     "name": "VIDEO",
+     "type": "VIDEO",
+     "links": [
+      21
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "CreateVideo"
+   },
+   "widgets_values": [
+    24,
+    8,
+    "sRGB",
+    "none"
+   ]
+  },
+  {
+   "id": 16,
+   "type": "SaveVideo",
+   "pos": [
+    1720,
+    0
+   ],
+   "size": [
+    520,
+    560
+   ],
+   "flags": {},
+   "order": 15,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "video",
+     "type": "VIDEO",
+     "link": 21
+    },
+    {
+     "name": "filename_prefix",
+     "type": "STRING",
+     "widget": {
+      "name": "filename_prefix"
+     },
+     "link": null
+    },
+    {
+     "name": "format",
+     "type": "COMFY_DYNAMICCOMBO_V3",
+     "widget": {
+      "name": "format"
+     },
+     "link": null
+    },
+    {
+     "name": "format.codec",
+     "type": "COMFY_DYNAMICCOMBO_V3",
+     "widget": {
+      "name": "format.codec"
+     },
+     "link": null
+    },
+    {
+     "name": "codec",
+     "shape": 7,
+     "type": "COMFY_DYNAMICCOMBO_V3",
+     "widget": {
+      "name": "codec"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "video",
+     "type": "VIDEO",
+     "links": []
+    }
+   ],
+   "properties": {
+    "cnr_id": "comfy-core",
+    "ver": "0.33.0",
+    "Node name for S&R": "SaveVideo"
+   },
+   "widgets_values": [
+    "video/dmad_h3_4step_5s_8gb_guitar",
+    "auto",
+    "auto",
+    "auto"
+   ]
+  },
+  {
+   "id": 18,
+   "type": "H3MemoryOptimization",
+   "pos": [
+    0,
+    800
+   ],
+   "size": [
+    430,
+    200
+   ],
+   "flags": {},
+   "order": 16,
+   "mode": 0,
+   "inputs": [
+    {
+     "name": "model",
+     "type": "MODEL",
+     "link": 3
+    },
+    {
+     "name": "fused_qkv",
+     "type": "COMBO",
+     "widget": {
+      "name": "fused_qkv"
+     },
+     "link": null
+    },
+    {
+     "name": "mlp_memory",
+     "type": "COMBO",
+     "widget": {
+      "name": "mlp_memory"
+     },
+     "link": null
+    },
+    {
+     "name": "chunk_rows",
+     "type": "INT",
+     "widget": {
+      "name": "chunk_rows"
+     },
+     "link": null
+    },
+    {
+     "name": "preserve_precision",
+     "type": "BOOLEAN",
+     "widget": {
+      "name": "preserve_precision"
+     },
+     "link": null
+    },
+    {
+     "name": "precision_mode",
+     "type": "COMBO",
+     "widget": {
+      "name": "precision_mode"
+     },
+     "link": null
+    },
+    {
+     "name": "qkv_streaming_mode",
+     "type": "COMBO",
+     "widget": {
+      "name": "qkv_streaming_mode"
+     },
+     "link": null
+    },
+    {
+     "name": "embedding_memory_mode",
+     "type": "COMBO",
+     "widget": {
+      "name": "embedding_memory_mode"
+     },
+     "link": null
+    },
+    {
+     "name": "kitchen_v_memory_mode",
+     "type": "COMBO",
+     "widget": {
+      "name": "kitchen_v_memory_mode"
+     },
+     "link": null
+    }
+   ],
+   "outputs": [
+    {
+     "name": "MODEL",
+     "type": "MODEL",
+     "links": [
+      4
+     ]
+    }
+   ],
+   "properties": {
+    "cnr_id": "h3-optimizations",
+    "ver": "0.2.45",
+    "Node name for S&R": "H3MemoryOptimization"
+   },
+   "widgets_values": [
+    "auto",
+    "auto",
+    1024,
+    true,
+    "Auto",
+    "Auto",
+    "Auto",
+    "Lower VRAM (slower)"
+   ]
+  },
+  {
+   "id": 17,
+   "type": "MarkdownNote",
+   "pos": [
+    480,
+    720
+   ],
+   "size": [
+    890,
+    330
+   ],
+   "flags": {},
+   "order": 18,
+   "mode": 0,
+   "inputs": [],
+   "outputs": [],
+   "properties": {},
+   "widgets_values": [
+    "## DMAD × MiniMax-H3 — 4-step text-to-audio-video (8 GB GPU)\n\n**Models** (put in `ComfyUI/models/...`):\n- `diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors` — [Comfy-Org/MiniMax-H3](https://huggingface.co/Comfy-Org/MiniMax-H3)\n- `text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors` — same repo\n- `vae/minimax_h3_video_vae_fp16.safetensors`, `vae/minimax_h3_audio_vae_fp32.safetensors` — same repo\n- `loras/dmad_minimax_h3_4step_full_critic_comfyui.safetensors` — [ZhengmingYu/DMAD](https://huggingface.co/ZhengmingYu/DMAD/tree/main/minimax_h3)\n\n**Settings that matter:** LoRA strength 1.0 · shift 12 / audio 2 · sampler `lcm` + scheduler `simple` · no CFG.\n**Steps:** 4 = fastest; 8 or 12 give finer detail on fast motion. Avoid `euler` (softer).\n**Prompt:** describe the shot, then `overall_soundscape:` and `non_diegetic_music:` for the audio.\n**8 GB GPUs:** 124 frames = 5 s at 24 fps. The `H3 Memory Optimization` node (Lower VRAM attention, 1024-row\nchunks) and the `H3 AIMDO Residency Limiter` (0 blocks) of [H3-Optimizations](https://github.com/Zironic/H3-Optimizations)\n(install with ComfyUI-Manager) make it fit in 8 GB (it runs with 6 GB free); the model streams from system memory (64 GB RAM\nrecommended).\nFor 15 s use the 4- or 8-step 15 s workflows on a 16 GB GPU or larger.\n"
+   ]
+  }
+ ],
+ "links": [
+  [
+   1,
+   1,
+   0,
+   2,
+   0,
+   "MODEL"
+  ],
+  [
+   2,
+   2,
+   0,
+   3,
+   0,
+   "MODEL"
+  ],
+  [
+   3,
+   3,
+   0,
+   18,
+   0,
+   "MODEL"
+  ],
+  [
+   4,
+   18,
+   0,
+   19,
+   0,
+   "MODEL"
+  ],
+  [
+   5,
+   19,
+   0,
+   9,
+   0,
+   "MODEL"
+  ],
+  [
+   6,
+   19,
+   0,
+   11,
+   0,
+   "MODEL"
+  ],
+  [
+   7,
+   4,
+   0,
+   7,
+   0,
+   "CLIP"
+  ],
+  [
+   8,
+   5,
+   0,
+   7,
+   1,
+   "VAE"
+  ],
+  [
+   9,
+   7,
+   0,
+   9,
+   1,
+   "CONDITIONING"
+  ],
+  [
+   10,
+   8,
+   0,
+   12,
+   0,
+   "NOISE"
+  ],
+  [
+   11,
+   9,
+   0,
+   12,
+   1,
+   "GUIDER"
+  ],
+  [
+   12,
+   10,
+   0,
+   12,
+   2,
+   "SAMPLER"
+  ],
+  [
+   13,
+   11,
+   0,
+   12,
+   3,
+   "SIGMAS"
+  ],
+  [
+   14,
+   7,
+   1,
+   12,
+   4,
+   "LATENT"
+  ],
+  [
+   15,
+   12,
+   0,
+   13,
+   0,
+   "LATENT"
+  ],
+  [
+   16,
+   5,
+   0,
+   13,
+   1,
+   "VAE"
+  ],
+  [
+   17,
+   12,
+   0,
+   14,
+   0,
+   "LATENT"
+  ],
+  [
+   18,
+   6,
+   0,
+   14,
+   1,
+   "VAE"
+  ],
+  [
+   19,
+   13,
+   0,
+   15,
+   0,
+   "IMAGE"
+  ],
+  [
+   20,
+   14,
+   0,
+   15,
+   1,
+   "AUDIO"
+  ],
+  [
+   21,
+   15,
+   0,
+   16,
+   0,
+   "VIDEO"
+  ]
+ ],
+ "groups": [],
+ "config": {},
+ "extra": {
+  "ds": {
+   "scale": 0.6,
+   "offset": [
+    40,
+    60
+   ]
+  },
+  "frontendVersion": "1.49.6"
+ },
+ "version": 0.4
+}

minimax_h3/workflows/dmad_h3_4step_5s_8gb_guitar.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5892a42e099472eaa001df409139f74a312bd8194557a78aa8d50af73152ed51
+size 2607452