This project provides an enhanced, unofficial implementation of the BiRefNet background removal model for ComfyUI, targeting users who need efficient and versatile image and video segmentation. It offers significant improvements over existing solutions by separating model loading and processing for speed, enabling direct output of transparent PNGs, and supporting video background removal.

How It Works

The implementation leverages the BiRefNet model, described as a top-tier open-source, commercially viable background removal solution. Key architectural choices include separating model loading from image processing, a design also seen in other ComfyUI nodes, which boosts performance. This allows for direct output of transparent PNGs and streamlined video segmentation.

Quick Start & Requirements

• Install via git clone https://github.com/ZHO-ZHO-ZHO/ComfyUI-BiRefNet-ZHO.git into the custom_nodes directory.
• Install dependencies: pip install -r requirements.txt (requires timm).
• Download the 6 BiRefNet models to ./models/BiRefNet.
• ComfyUI Manager installation is planned.
• Official documentation and demo videos are available via GitHub links.

Highlighted Details

• Supports both image and video processing, including batch processing.
• Directly outputs transparent background PNG images.
• Enables background removal for video files.
• Model loading and image processing are separated for improved speed.

Maintenance & Community

• The project is maintained by ZHO-ZHO-ZHO.
• A QQ group (839821928) is available for community support.
• Links to Bilibili and X (Twitter) are provided.

Licensing & Compatibility

• The README does not explicitly state a license. The project references viperyl/ComfyUI-BiRefNet, which is MIT licensed. Clarification on licensing is recommended for commercial use.

Limitations & Caveats

The project is an unofficial implementation and does not explicitly state its license, which may impact commercial adoption. Installation via ComfyUI Manager is still pending.