rembg removes the background from images with pretrained segmentation models (U2-Net, BiRefNet, BRIA RMBG, SAM and others) with ONNX Runtime. It processes single files, whole folders, binary streams, or serves an HTTP API with a browser interface. CPU support is included; NVIDIA GPU support is available through the optional cuda component.

Examples

rembg i ~/Pictures/photo.jpg ~/Pictures/photo-cut.png
rembg i -m birefnet-general -a ~/Pictures/photo.jpg out.png
rembg p ~/Pictures/input ~/Pictures/output
cat photo.jpg | rembg i > photo-cut.png
rembg s
rembg d bria-rmbg

Models

Models are downloaded on first use and kept in ~/snap/rembg/common/models, which survives snap refreshes. Custom models passed with model_path must be placed under ~/snap/rembg/common.

File access

rembg can use images in your home directory. To also use USB drives and other removable media (/media, /mnt), run:

sudo snap connect rembg:removable-media