Render-AI / MegFaceAnimate

MegActor: Harness the Power of Raw Video for Vivid Portrait Animation

Shurong yang^*, Huadong Li^*, Juhao Wu^*, Minhao Jing^*†, Linze Li, Renhe Ji^‡, Jiajun Liang^‡, Haoqiang Fan

MEGVII Technology

^*Equal contribution ^†Lead this project ^‡Corresponding author

News & TODO List

• [✅2024.05.24] Inference settings are released.

• [❌] Data curation pipeline to be released .

• [❌] Training setup to be released.

Overview

MegActor is an intermediate-representation-free portrait animator that uses the original video, rather than intermediate features, as the driving factor to generate realistic and vivid talking head videos. Specifically, we utilize two UNets: one extracts the identity and background features from the source image, while the other accurately generates and integrates motion features directly derived from the original videos. MegActor can be trained on low-quality, publicly available datasets and excels in facial expressiveness, pose diversity, subtle controllability, and visual quality.

Pre-generated results

demo.mp4

demo4.mp4

demo6.mp4

Preparation

• Environments

  Detailed environment settings should be found with environment.yaml

  conda env create -f environment.yaml
  pip install -U openmim
  
  mim install mmengine
  mim install "mmcv>=2.0.1"
  mim install "mmdet>=3.1.0"
  mim install "mmpose>=1.1.0"
  
  conda install -c conda-forge cudatoolkit-dev -y
  

• Dataset

  To be released.

• Pretrained weights

  Please find our pretrained weights at https://huggingface.co/HVSiniX/RawVideoDriven. Or simply use

  git clone https://huggingface.co/HVSiniX/RawVideoDriven && ln -s RawVideoDriven/weights weights

Training

To be released.

Inference

Currently only single-GPU inference is supported.

CUDA_VISIBLE_DEVICES=0 python eval.py --config configs/infer12_catnoise_warp08_power_vasa.yaml --source {source image path} --driver {driving video path}

Demo

For gradio interface, please run

python demo/run_gradio.py

Acknowledgement

Many thanks to the authors of mmengine, MagicAnimate, Controlnet_aux, and Detectron2.

Contact

If you have any questions, feel free to open an issue or contact us at [e-mail], [e-mail] or [e-mail].