rembg

A local AI background removal tool using ONNX models, with CPU, NVIDIA and AMD GPU support and a self-hosted HTTP server.

Alternative to:

rembg removes image backgrounds on your own hardware for cutouts and automated image workflows. Its local ONNX models keep image processing on your machine after the initial model download.

Workflow integration: A command-line tool, Python library, HTTP server and Docker support let you add background removal to scripts or self-hosted services.
Model choice: The default bria-rmbg model produces softer edges but runs more slowly than u2net; specialist choices include birefnet-portrait for portraits and isnet-anime for anime characters.
Edge refinement: Color decontamination reduces halos around soft boundaries, while alpha matting and ViTMatte also refine transparency at a higher processing cost.

rembg supports CPU processing, compatible NVIDIA GPUs through CUDA and AMD GPUs through ROCm. Model downloads require an internet connection; the optional withoutBG backend requires an API key and sends images to withoutBG's cloud servers instead of using a local model.

The software is open source under MIT, but model weights have separate licenses. The default BRIA model requires a paid agreement for commercial use.