2025-08-13 22:14:54 +08:00
I
2025-08-13 16:07:09 +08:00
2025-08-13 16:05:03 +08:00
2025-08-13 22:14:54 +08:00

ComfyUI_StableAvatar

StableAvatar: Infinite-Length Audio-Driven Avatar Video Generation,you can try it in ComfyUI

UPDATE

  • Coming soon,特点,无限长(音频多长就推理多长)

1. Installation

In the ./ComfyUI/custom_nodes directory, run the following:

git clone https://github.com/smthemex/ComfyUI_StableAvatar.git

2. Requirements

pip install -r requirements.txt

3.models

3.1 fromFrancisRing/StableAvatar downlaod "Wan2.1_VAE.pth" ,"diffusion_pytorch_model.safetensors" and "config.json ","Kim_Vocal_2.onnx" ,"transformer3d-rec-vec.pt" or "transformer3d-square.pt " 底模有2个可选
3.2 use comfyui ,clipvison-h and umt5_xxl_fp8_e4m3fn_scaled.safetensors
3.3 wav2vec2-base-960h
3.4 if use echomimic v3,just only download"transformer3d-rec-vec.pt" or "transformer3d-square.pt " and "Kim_Vocal_2.onnx" / 如果也用echomimic v3,仅需下载底模和Kim_Vocal_2.onnx,会自动调用echomimic的模型

├── ComfyUI/models/StableAvatar/transformer 
|         ├── diffusion_pytorch_model.safetensors  # Wan2.1-Fun-V1.1-1.3B-InP transformer #3.13G 务必注意模型同名。
|         ├── config.json
├── ComfyUI/models/StableAvatar/wav2vec2-base-960h
|         ├── all config json files 
|         ├──  model.safetensors
├── ComfyUI/models/clip
|         ├── umt5_xxl_fp8_e4m3fn_scaled.safetensors #comfy
├── ComfyUI/models/clip_vision
|         ├──clipvison-h # 1.26G comfy
├── ComfyUI/models/diffusion_models/
|         ├──transformer3d-rec-vec.pt  # FrancisRing/StableAvatar 二选一
|         ├──transformer3d-square.pt   # FrancisRing/StableAvatar
├── ComfyUI/models/vae
|         ├── Wan2.1_VAE.pth
├── ComfyUI/models/StableAvatar/  # 音频分离用
|         ├──Kim_Vocal_2.onnx

4.Tips

  • 480x832 or 832x480 or 512x512 resolution /模型在这三种分辨率训练,推荐使用
  • overlap_window_length 越大越好越慢/ high will get best quality but more times
  • step 30~50
  • 二种gpu卸载方式

5. Example

6 .Citation

@article{tu2025stableavatar,
  title={StableAvatar: Infinite-Length Audio-Driven Avatar Video Generation},
  author={Tu, Shuyuan and Pan, Yueming and Huang, Yinming and Han, Xintong and Xing, Zhen and Dai, Qi and Luo, Chong and Wu, Zuxuan and Jiang Yu-Gang},
  journal={arXiv preprint arXiv:2508.08248},
  year={2025}
}
S
Description
No description provided
Readme MIT
146 MiB
Languages
Python 99.1%
Shell 0.9%