ComfyUI Extension: ComfyUI_BiRefNet_ll

ComfyUI_BiRefNet_ll Introduction

ComfyUI_BiRefNet_ll is an extension designed to enhance the capabilities of the ComfyUI by integrating the BiRefNet model for high-resolution dichotomous image segmentation. This extension allows AI artists to easily remove backgrounds from images, making it an invaluable tool for creating clean and professional visuals. Whether you're working on general images, portraits, or specific tasks like concealed object detection, ComfyUI_BiRefNet_ll provides a robust solution to streamline your workflow.

How ComfyUI_BiRefNet_ll Works

ComfyUI_BiRefNet_ll leverages the power of the BiRefNet model, which is designed for high-resolution image segmentation. Think of it as a highly skilled digital pair of scissors that can precisely cut out objects from their backgrounds. The model uses advanced algorithms to distinguish between the foreground (the main subject) and the background, allowing you to isolate the subject with high accuracy. This process involves analyzing the image at a pixel level to create a mask that separates the foreground from the background.

ComfyUI_BiRefNet_ll Features

AutoDownloadBiRefNetModel

• Description: Automatically downloads the required BiRefNet models to the models/BiRefNet directory.
• Customization: This feature does not support downloading weight models.

LoadRembgByBiRefNetModel

• Description: Loads the BiRefNet model from the models/BiRefNet directory or a specified path in an extra YAML file.
• Customization: Supports both new and old BiRefNet models. When the use_weight parameter is set to True, it requires downloading the weight models.

RembgByBiRefNet

• Description: This node is used to perform the actual background removal using the BiRefNet model.
• Customization: You can choose different models and adjust parameters to fine-tune the background removal process.

ComfyUI_BiRefNet_ll Models

Latest Models:

• General: A pre-trained model for general use cases.
• General-Lite: A lightweight version of the General model for faster performance.
• Portrait: Specifically trained for human portraits.
• DIS: Designed for dichotomous image segmentation.
• HRSOD: High-resolution salient object detection.
• COD: Concealed object detection.
• DIS-TR_TEs: A model trained on a massive dataset for enhanced performance.

Old Models:

• BiRefNet-DIS_ep580.pth: An older version of the DIS model.
• BiRefNet-ep480.pth: Another older model for general use.

Model Download Links:

• General ➔ Rename to General.safetensors
• General-Lite ➔ Rename to General-Lite.safetensors
• Portrait ➔ Rename to Portrait.safetensors
• DIS ➔ Rename to DIS.safetensors
• HRSOD ➔ Rename to HRSOD.safetensors
• COD ➔ Rename to COD.safetensors
• DIS-TR_TEs ➔ Rename to DIS-TR_TEs.safetensors

Troubleshooting ComfyUI_BiRefNet_ll

Common Issues and Solutions

1. Model Not Found:

• Solution: Ensure that the model files are correctly named and placed in the models/BiRefNet directory. If using AutoDownloadBiRefNetModel, verify that the download process completed successfully.

2. Background Removal Not Accurate:

• Solution: Try using a different model that better suits your image type. For example, use the Portrait model for human subjects.

3. Slow Performance:

• Solution: Use the General-Lite model for faster performance, especially on lower-end hardware.

Frequently Asked Questions

• Q: Can I use this extension for video background removal?
• A: Currently, ComfyUI_BiRefNet_ll is optimized for image background removal. For video, consider using specialized video editing software.
• Q: Do I need a powerful GPU to use this extension?
• A: While a powerful GPU can speed up the process, the General-Lite model is designed to work efficiently on less powerful hardware.

Learn More about ComfyUI_BiRefNet_ll

For additional resources, tutorials, and community support, you can explore the following links:

• BiRefNet Project Page
• Hugging Face Model
• Colab Demo for Multiple Images Inference
• Colab Demo for Inference and Evaluation
• Colab Demo for Box-Guided Segmentation These resources will help you get the most out of ComfyUI_BiRefNet_ll and connect with other users and developers in the community.