ComfyUI Extension: ComfyUI_BiRefNet_ll

中文文档

Support the use of new and old versions of BiRefNet models

Preview

Install

• Manual

    cd custom_nodes
    git clone https://github.com/lldacing/ComfyUI_BiRefNet_ll.git
    cd ComfyUI_BiRefNet_ll
    pip install -r requirements.txt
    # restart ComfyUI

• Via ComfyUI Manager

Models

The available newest models are:

• General: A pre-trained model for general use cases.
• General-HR: A pre-trained model for general use cases which shows great performance on higher resolution images (2048x2048).
• General-Lite: A light pre-trained model for general use cases.
• General-Lite-2K: A light pre-trained model for general use cases in high resolution (2560x1440).
• General-dynamic: A pre-trained model for dynamic resolution, trained with images in range from 256x256 to 2304x2304.
• General-reso_512: A pre-trained model for faster and more accurate lower resolution, trained with images in 512x512.
• General-legacy: A pre-trained model for general use trained on DIS5K-TR,DIS-TEs, DUTS-TR_TE,HRSOD-TR_TE,UHRSD-TR_TE, HRS10K-TR_TE (w/o portrait seg data).
• Portrait: A pre-trained model for human portraits.
• Matting: A pre-trained model for general trimap-free matting use.
• Matting-HR: A pre-trained model for general matting use which shows great matting performance on higher resolution images (2048x2048).
• Matting-Lite: A light pre-trained model for general trimap-free matting use.
• DIS: A pre-trained model for dichotomous image segmentation (DIS).
• HRSOD: A pre-trained model for high-resolution salient object detection (HRSOD).
• COD: A pre-trained model for concealed object detection (COD).
• DIS-TR_TEs: A pre-trained model with massive dataset.

Model files go here (when use AutoDownloadBiRefNetModel automatically downloaded if the folder is not present during first run): ${comfyui_rootpath}/models/BiRefNet.

If necessary, they can be downloaded from:

• General ➔ model.safetensors must be renamed General.safetensors
• General-HR ➔ model.safetensors must be renamed General-HR.safetensors
• General-Lite ➔ model.safetensors must be renamed General-Lite.safetensors
• General-Lite-2K ➔ model.safetensors must be renamed General-Lite-2K.safetensors
• General-dynamic ➔ model.safetensors must be renamed General-dynamic.safetensors
• General-legacy ➔ model.safetensors must be renamed General-legacy.safetensors
• General-reso_512 ➔ model.safetensors must be renamed General-reso_512.safetensors
• Portrait ➔ model.safetensors must be renamed Portrait.safetensors
• Matting ➔ model.safetensors must be renamed Matting.safetensors
• Matting-HR ➔ model.safetensors must be renamed Matting-HR.safetensors
• Matting-Lite ➔ model.safetensors must be renamed Matting-Lite.safetensors
• DIS ➔ model.safetensors must be renamed DIS.safetensors
• HRSOD ➔ model.safetensors must be renamed HRSOD.safetensors
• COD ➔ model.safetensors must be renamed COD.safetensors
• DIS-TR_TEs ➔ model.safetensors must be renamed DIS-TR_TEs.safetensors

Some models on GitHub: BiRefNet Releases

Old models:

• BiRefNet-DIS_ep580.pth
• BiRefNet-ep480.pth

Weight Models (Optional)

• swin_large_patch4_window12_384_22kto1k.pth(not General-Lite, General-Lite-2K and Matting-Lite model)
• swin_tiny_patch4_window7_224_22kto1k_finetune.pth(just General-Lite, General-Lite-2K and Matting-Lite model)

Nodes

• AutoDownloadBiRefNetModel
  • Automatically download the model into ${comfyui_rootpath}/models/BiRefNet, do not support weight model
• LoadRembgByBiRefNetModel
  • Can select model from ${comfyui_rootpath}/models/BiRefNet or the path of birefnet configured in the extra YAML file
  • You can download latest models from BiRefNet Releases or old models BiRefNet-DIS_ep580.pth and BiRefNet-ep480.pth
  • When param use_weight is True, need download weight model swin_large_patch4_window12_384_22kto1k.pth
    model General-Lite, General-Lite-2K and Matting-Lite must use weight model swin_tiny_patch4_window7_224_22kto1k_finetune.pth
• RembgByBiRefNet
  • Output transparent foreground image and mask
• RembgByBiRefNetAdvanced
  • Output foreground image and mask, provide some fine-tuning parameters
• GetMaskByBiRefNet
  • Only output mask
• BlurFusionForegroundEstimation
  • Use the fast-foreground-estimation method to estimate the foreground image

Thanks

ZhengPeng7/BiRefNet

dimitribarbot/sd-webui-birefnet