Viggle-Animate (MiniMax H3) 12GB — character replacement in video, 4 steps, the driving clip's audio kept
10
141
13
☕ One coffee = one happy creator: https://buymeacoffee.com/the_frizzy1
Swap the person in any video for your character from one still. The motion, the camera, the background and the driving clip's audio all stay, in 4 steps. There is no video on this workflow yet: the one on line 1 is how to use my workflows, if you are new to them.
What you need
✔ Custom nodes — comfyui-viggle-animate-h3 from the Comfy registry, v1.3.2. Everything else is core ComfyUI
✔ ComfyUI version — 0.37.4 or newer
✔ VRAM — 12 GB — tested, not estimated. Peak 11.2–11.6 GB
✔ System RAM — 32 GB tested · 14.7–15.9 GB peak in the runs below
✔ Tested on — RTX 3060 12 GB · 32 GB DDR4 · ComfyUI 0.37.4 (Sage attention)
✔ Licence — Workflow MIT · Viggle-Animate: MiniMax H3 Community License
Models to download
Click a file name to download it. Put it in the folder shown, inside ComfyUI/models/. Every file name was read from the workflow .json.
Required — 4 files.
diffusion_models/
File: minimax_h3_ref2va_viggle_pruned_int8_convrot.safetensors — 21.0 GB (12 GB (tested))
Download:
https://huggingface.co/drbaph/Viggle-Animate-ComfyUI/resolve/main/diffusion_models/minimax_h3_ref2va_viggle_pruned_int8_convrot.safetensors
Put it here:
ComfyUI/models/diffusion_models/
loras/
File: viggle_animate_dmd_lora_r64.safetensors — 0.9 GB
Download:
https://huggingface.co/drbaph/Viggle-Animate-ComfyUI/resolve/main/loras/viggle_animate_dmd_lora_r64.safetensors
Put it here:
ComfyUI/models/loras/
text_cond/
File: fixed_embed_fwd_anyframe.safetensors — 0.0 GB
Download:
https://huggingface.co/drbaph/Viggle-Animate-ComfyUI/resolve/main/text_cond/fixed_embed_fwd_anyframe.safetensors
Put it here:
ComfyUI/models/text_cond/
Comfy-Org/MiniMax-H3 — note this is the fp16 video VAE, not the int8 one the other H3 workflows use.
vae/
File: minimax_h3_video_vae_fp16.safetensors — 5.2 GB
Download:
https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_fp16.safetensors
Put it here:
ComfyUI/models/vae/
What it does
One mode and one mode only: character replacement, a driving video plus one character still. It is Viggle's 33B finetune of MiniMax H3 Ref2VA, run with its node pack's conditioning and loader nodes and core ComfyUI for everything else: native video load and save instead of VHS, ManualSigmas instead of custom sigma nodes. The driving clip's audio is kept.
Load The_frizzy1_viggle-animate-h3_v2.0.0.json, pick a mode, add your inputs, Run. If your models live in sub-folders the loaders show red until you re-pick them once.
Speed on my RTX 3060
Each run went once through this exact workflow, 5 s clips. Wall is queue to finished including model loading, sampling is time in the sampler nodes.
Doll street-dance clip — 480x864, 4 min 33 s wall, 2 min 49 s sampling, 11.3 GB peak vram, 15.8 GB peak ram
Talking clip, landscape — 832x480, 4 min 00 s wall, 2 min 36 s sampling, 11.6 GB peak vram, 15.9 GB peak ram
Talking clip, portrait — 480x832, 3 min 53 s wall, 2 min 33 s sampling, 11.2 GB peak vram, 14.7 GB peak ram
What did not work
The README has no known-issues section, so there is no tested failure list for this one yet. What the numbers do say: peak VRAM sits at 11.2 to 11.6 GB on a 12 GB card, so it barely fits, and that is on 5 s clips at 480x864 or 832x480. Longer or bigger is untested here. This is also the one workflow of mine that needs a custom node pack rather than core nodes only.
Get the workflow
GitHub: https://github.com/Thefrizzy1/The_frizzy1-Workflows/tree/main/workflows/minimax/viggle-animate-h3
Hugging Face mirror: https://huggingface.co/The-frizzy1/The_frizzy1-Workflows/tree/main/workflows/minimax/viggle-animate-h3
More videos on this stuff: https://www.youtube.com/playlist?list=UUGKDBMhTGqvlBZE9cF5d50Q