Open background removal models#
Pick a model for the job: license, rembg session, ONNX footprint, and side-by-side evidence on the same Bench v0 images — including MAE / edge / connectivity vs ground-truth mattes.
Product install and docs: withoutBG Open Model hub.
Our open model
withoutBG Open Weights
Apache-2.0 open-weights background removal model from withoutBG. Self-host via Python, Docker, or Mac. Apache-2.0 · self-host with Python, Docker, Mac, or GIMP. Eval card and peer compares live here; install and docs live on the product hub.
Other open models
U²-Net
Apache-2.0Classic salient-object / background-removal network. Default model in many rembg setups.
U²-Net · rembg `u2net` · ONNX · vs withoutBG Open
U²-NetP
Apache-2.0Lightweight U²-Net variant for faster CPU inference.
U²-Net · rembg `u2netp` · ONNX · vs withoutBG Open
Silueta
Apache-2.0Compact silhouette-oriented model shipped with rembg.
U²-Net · rembg `silueta` · ONNX · vs withoutBG Open
IS-Net (general use)
Apache-2.0High-resolution dichotomous image segmentation model (DIS / IS-Net) for general cutouts.
IS-Net · rembg `isnet-general-use` · ONNX · vs withoutBG Open
BiRefNet General
MITBilateral Reference Network for high-quality dichotomous segmentation / matting-style cutouts.
BiRefNet · rembg `birefnet-general` · ONNX · vs withoutBG Open
BiRefNet General Lite
MITLighter BiRefNet general checkpoint for faster inference.
BiRefNet · rembg `birefnet-general-lite` · ONNX · vs withoutBG Open
BRIA RMBG
CustomBRIA background-removal model packaged for rembg (check license for commercial use).
BRIA · rembg `bria-rmbg` · ONNX · vs withoutBG Open
Best for
- Best Open Models for Product Photography
- Best Open Models for Hair and Portraits
- Fastest Small Open Background Removal Models
Also see the comparison matrix and benchmark hub.