Awesome

Face Adapter for Pre-Trained Diffusion Models with Fine-Grained ID and Attribute Control

<a href='https://faceadapter.github.io/face-adapter.github.io/'><img src='https://img.shields.io/badge/Project-Page-green'></a> <a href='https://huggingface.co/FaceAdapter/FaceAdapter'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue'></a> <a href='https://huggingface.co/spaces/FaceAdapter/FaceAdapter'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue'></a>

Introduction

Face-Adapter is an efficient and effective face editing adapter for pre-trained diffusion models, specifically targeting face reenactment and swapping tasks.

Release

• [2024/5/25] 🔥 We release the gradio demo.
• [2024/5/24] 🔥 We release the code and models.

Installation

# Torch >= 2.0 recommended for acceleration without xformers
pip install accelerate diffusers==0.26.0 insightface onnxruntime

Download Models

You can download models of FaceAdapter directly from here or download using python script:

# Download all files 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="FaceAdapter/FaceAdapter", local_dir="./checkpoints")

# If you want to download one specific file
from huggingface_hub import hf_hub_download
hf_hub_download(repo_id="FaceAdapter/FaceAdapter", filename="controlnet/config.json", local_dir="./checkpoints")

To run the demo, you should also download the pre-trained SD models below:

• runwayml/stable-diffusion-v1-5
• stabilityai/sd-vae-ft-mse

⚡ Quick Inference

SD_1.5

python infer.py 

You can adjust the cropping size with the --crop_ratio （default:0.81）parameter. But be careful not to set the crop range too large, as this can decrease the quality of the generated images due to the limit of the training data size.

😊 FaceAdapter can be seamlessly plugged into community models:

python infer.py --base_model "frankjoshua/toonyou_beta6"

Disclaimer

This project strives to positively impact the domain of AI-driven image generation. Users are granted the freedom to create images using this tool, but they are expected to comply with local laws and utilize it in a responsible manner. The developers do not assume any responsibility for potential misuse by users.

Citation

If you find Face-Adapter useful for your research and applications, please cite using this BibTeX:

@article{han2024face,
  title={Face Adapter for Pre-Trained Diffusion Models with Fine-Grained ID and Attribute Control},
  author={Han, Yue and Zhu, Junwei and He, Keke and Chen, Xu and Ge, Yanhao and Li, Wei and Li, Xiangtai and Zhang, Jiangning and Wang, Chengjie and Liu, Yong},
  journal={arXiv preprint arXiv:2405.12970},
  year={2024}
}