ComfyUI-IDM-VTON

ComfyUI-IDM-VTON
★ 575

虚拟试衣服装迁移ComfyUI 插件高显存需求
将 a/IDM-VTON 的虚拟试衣功能集成到 ComfyUI，实现服装到人体的准确迁移，便于快速试穿；需 ≥16GB 显存。
💡 在 ComfyUI 中将服装样式转移到人物以实现虚拟试穿。
🍴 66 Forks💻 Python🔄 2024-08-20
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/8b0992d318c3
📦 requirements.txt
torch
torchvision
torchaudio
accelerate==0.30.0
torchmetrics==1.4.0
tqdm==4.66.4
transformers==4.40.2
diffusers==0.27.2
einops==0.8.0
bitsandbytes==0.42
scipy==1.13.0
opencv-python
workflow
📄 README

ComfyUI-IDM-VTON

ComfyUI adaptation of IDM-VTON.

Installation

:warning: Current implementation requires GPU with at least 16GB of VRAM :warning:

Using ComfyUI Manager:

  • In ComfyUI Manager, look for “ComfyUI-IDM-VTON`, and be sure the author is `TemryL“. Install it.

    • Manually:

  • Clone this repo into custom_nodes folder in ComfyUI and install the dependencies.
    • cd custom_nodes
git clone https://github.com/TemryL/ComfyUI-IDM-VTON.git
cd ComfyUI-IDM-VTON
python install.py

    Models weights from yisol/IDM-VTON in HuggingFace will be downloaded in models folder of this repository.

    Mask Generation

    The workflow provided above uses ComfyUI Segment Anything to generate the image mask.

    DensePose Estimation

    DensePose estimation is performed using ComfyUI’s ControlNet Auxiliary Preprocessors.

    :star: Star Us!

    If you find this project useful, please consider giving it a star on GitHub. This helps the project to gain visibility and encourages more contributors to join in. Thank you for your support!

    Contribute

    Thanks for your interest in contributing to the source code! We welcome help from anyone and appreciate every contribution, no matter how small!

    If you’re ready to contribute, please create a fork, make your changes, commit them, and then submit a pull request for review. We’ll consider it for integration into the main code base.

    Credits

  • ComfyUI
  • IDM-VTON
  • ComfyUI Segment Anything
  • ComfyUI’s ControlNet Auxiliary Preprocessors

    • License

    Original IDM-VTON source code under CC BY-NC-SA 4.0 license.