1038lab/ComfyUI-RMBG

What it solves

ComfyUI-RMBG provides a comprehensive suite of tools for removing image backgrounds and performing precise object segmentation. It eliminates the need for manual masking by automating the separation of subjects from their backgrounds, specifically targeting complex elements like faces, clothing, fashion items, and transparent objects.

How it works

The project implements a collection of custom nodes for ComfyUI that integrate multiple state-of-the-art AI models. It supports a wide range of models including RMBG-2.0, BiRefNet, SAM (1, 2, and 3), Florence-2, and YOLOv8. Users can perform segmentation via text prompts (natural language or tags) or use specialized models for specific categories like fashion or facial features.

Who it’s for

Digital artists, AI image generators, and designers using ComfyUI who need high-quality, automated masking and background removal for their workflows.

Highlights

• Diverse Model Support: Integrates a vast array of models including SAM3 Multiplex, BiRefNet, and SDMatte for various segmentation tasks.
• Specialized Segmentation: Dedicated nodes for face parsing (19 categories) and clothing/fashion segmentation (18 categories).
• Text-Prompted Masking: Ability to generate masks based on natural language descriptions using SAM2 and GroundingDINO.
• ** uma VRAM Optimization**: Features an unload_model toggle for instant GPU memory release and native Apple Silicon (MPS) acceleration.
• Edge Refinement: Includes foreground refinement and enhanced edge detection for better transparency and detail preservation.