SDMatte - SafeTensors Models for Interactive Matting
This repository provides SafeTensors versions of the SDMatte models for interactive image matting, optimized for seamless use with ComfyUI.
🔍 About SDMatte
SDMatte: Grafting Diffusion Models for Interactive Matting is a state-of-the-art model that leverages the power of diffusion priors to achieve high-precision matting — especially around fine details and complex edges.
✨ Key Features
- Diffusion-Powered: Uses strong priors from diffusion models to extract high-fidelity details
- Interactive Matting: Visual prompt-driven control for intuitive editing
- Edge & Texture Focus: Excels in handling challenging edge regions and fine textures
- Coordinate & Opacity Awareness: Improves matting accuracy with spatial and opacity context
📦 Available Models
SDMatte.safetensors– Standard version for interactive mattingSDMatte_plus.safetensors– Enhanced version with improved performance
🧩 Built for ComfyUI: ComfyUI-RMBG
These models are designed for use with our ComfyUI custom node:
➡️ ComfyUI-RMBG on GitHub
This custom node integrates SDMatte into ComfyUI workflows, enabling high-quality interactive matting inside a visual pipeline.
🔄 Latest Update
Version: v2.9.0
Date: 2025-08-18
📄 Read the update changelog
🙌 Credits and Attribution
📚 Original Work
- Authors: vivoCameraResearch Team
- Model Repository: Hugging Face – LongfeiHuang/SDMatte
- Official Code: GitHub – vivoCameraResearch/SDMatte
- Paper: SDMatte: Grafting Diffusion Models for Interactive Matting
📝 Abstract (from the original paper)
Recent interactive matting methods have shown satisfactory performance in capturing the primary regions of objects, but they fall short in extracting fine-grained details in edge regions. Diffusion models trained on billions of image-text pairs demonstrate exceptional capability in modeling highly complex data distributions and synthesizing realistic texture details, while exhibiting robust text-driven interaction capabilities — making them an attractive solution for interactive matting.