FluxFit: Virtual Fitting based on Flux

· The first work to extend the powerful Flux model to virtual fitting task. It can only be used for Non-commercial Use.
· The effect is not ideal for some complex textured clothing and model with complex limbs, we are still trying to optimize it.

Updates

2024/10/12: Our Online Demo with v1.0 checkpoint is released.

Web Demo

Running FluxFit consumes a lot of GPU, we recommend using the online demo we provide to experience it.

Running on local

Download checkpoints for human parsing and dwpose from here and put under checkpoints folder.

checkpoints
|-- dwpose
    |-- dw-ll_ucoco_384.onnx
    |-- yolox_l.onnx
|-- humanparsing
    |-- parsing_atr.onnx
    |-- parsing_lip.onnx

You should first get access to FLUX.1-dev and login your account with huggingface-cli login

Chinese users can refer to this link to speed up model downloads.

Run demo with following command, requiring a gpu that supports bf16.

pip install diffusers==0.30.3

python gradio_flux.py # need about 66G GPU memory

python gradio_flux.py --offload # need about 32G GPU memory, the speed is slower

python gradio_flux.py --aggressive_offload # need about 8G GPU memory, the speed is slowest

Star History

License

The codes and checkpoints in this repository are under the CC BY-NC 4.0 license.

Contact

The model is developed by Boyuan Jiang at Tencent Youtu Lab. If you are interested in collaboration, feel free to contact me.

BoyuanJiang / FluxFit

readme