easonyan / aniportrait

AniPortrait: Audio-Driven Synthesis of Photorealistic Portrait Animation

AniPortrait

AniPortrait: Audio-Driven Synthesis of Photorealistic Portrait Animations

Author: Huawei Wei, Zejun Yang, Zhisheng Wang

Organization: Tencent Games Zhiji, Tencent

Here we propose AniPortrait, a novel framework for generating high-quality animation driven by audio and a reference portrait image. You can also provide a video to achieve face reenacment.

Pipeline

Updates / TODO List

• ✅ [2024/03/27] Now our paper is available on arXiv.

• ✅ [2024/03/27] Update the code to generate pose_temp.npy for head pose control.

• ✅ [2024/04/02] Update a new pose retarget strategy for vid2vid. Now we support substantial pose difference between ref_image and source video.

• ✅ [2024/04/03] We release our Gradio demo on HuggingFace Spaces (thanks to the HF team for their free GPU support)!

• 🔲 We will release audio2pose pre-trained weight for audio2video after futher optimization. You can choose head pose template in ./configs/inference/head_pose_temp as substitution.

Various Generated Videos

Self driven

cxk.mp4

solo.mp4

Face reenacment

num18.mp4

Aragaki.mp4

Video Source: 鹿火CAVY from bilibili

Audio driven

jijin.mp4	kara.mp4
lyl.mp4	zl.mp4

Installation

Build environment

We recommend a python version >=3.10 and cuda version =11.7. Then build environment as follows:

pip install -r requirements.txt

Download weights

All the weights should be placed under the ./pretrained_weights direcotry. You can download weights manually as follows:

1. Download our trained weights, which include four parts: denoising_unet.pth, reference_unet.pth, pose_guider.pth, motion_module.pth and audio2mesh.pt.

2. Download pretrained weight of based models and other components:

   • StableDiffusion V1.5
   • sd-vae-ft-mse
   • image_encoder
   • wav2vec2-base-960h

Finally, these weights should be orgnized as follows:

./pretrained_weights/
|-- image_encoder
|   |-- config.json
|   `-- pytorch_model.bin
|-- sd-vae-ft-mse
|   |-- config.json
|   |-- diffusion_pytorch_model.bin
|   `-- diffusion_pytorch_model.safetensors
|-- stable-diffusion-v1-5
|   |-- feature_extractor
|   |   `-- preprocessor_config.json
|   |-- model_index.json
|   |-- unet
|   |   |-- config.json
|   |   `-- diffusion_pytorch_model.bin
|   `-- v1-inference.yaml
|-- wav2vec2-base-960h
|   |-- config.json
|   |-- feature_extractor_config.json
|   |-- preprocessor_config.json
|   |-- pytorch_model.bin
|   |-- README.md
|   |-- special_tokens_map.json
|   |-- tokenizer_config.json
|   `-- vocab.json
|-- audio2mesh.pt
|-- denoising_unet.pth
|-- motion_module.pth
|-- pose_guider.pth
`-- reference_unet.pth

Note: If you have installed some of the pretrained models, such as StableDiffusion V1.5, you can specify their paths in the config file (e.g. ./config/prompts/animation.yaml).

Inference

Here are the cli commands for running inference scripts:

Kindly note that you can set -L to the desired number of generating frames in the command, for example, -L 300.

Self driven

python -m scripts.pose2vid --config ./configs/prompts/animation.yaml -W 512 -H 512

You can refer the format of animation.yaml to add your own reference images or pose videos. To convert the raw video into a pose video (keypoint sequence), you can run with the following command:

python -m scripts.vid2pose --video_path pose_video_path.mp4

Face reenacment

python -m scripts.vid2vid --config ./configs/prompts/animation_facereenac.yaml -W 512 -H 512

Add source face videos and reference images in the animation_facereenac.yaml.

Audio driven

python -m scripts.audio2vid --config ./configs/prompts/animation_audio.yaml -W 512 -H 512

Add audios and reference images in the animation_audio.yaml.

You can use this command to generate a pose_temp.npy for head pose control:

python -m scripts.generate_ref_pose --ref_video ./configs/inference/head_pose_temp/pose_ref_video.mp4 --save_path ./configs/inference/head_pose_temp/pose.npy

Training

Data preparation

Download VFHQ and CelebV-HQ

Extract keypoints from raw videos and write training json file (here is an example of processing VFHQ):

python -m scripts.preprocess_dataset --input_dir VFHQ_PATH --output_dir SAVE_PATH --training_json JSON_PATH

Update lines in the training config file:

data:
  json_path: JSON_PATH

Stage1

Run command:

accelerate launch train_stage_1.py --config ./configs/train/stage1.yaml

Stage2

Put the pretrained motion module weights mm_sd_v15_v2.ckpt (download link) under ./pretrained_weights.

Specify the stage1 training weights in the config file stage2.yaml, for example:

stage1_ckpt_dir: './exp_output/stage1'
stage1_ckpt_step: 30000 

Run command:

accelerate launch train_stage_2.py --config ./configs/train/stage2.yaml

Acknowledgements

We first thank the authors of EMO, and part of the images and audios in our demos are from EMO. Additionally, we would like to thank the contributors to the Moore-AnimateAnyone, majic-animate, animatediff and Open-AnimateAnyone repositories, for their open research and exploration.

Citation

@misc{wei2024aniportrait,
      title={AniPortrait: Audio-Driven Synthesis of Photorealistic Portrait Animations}, 
      author={Huawei Wei and Zejun Yang and Zhisheng Wang},
      year={2024},
      eprint={2403.17694},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}