Remove backgrounds. Free. Forever.

The tool runs a U²-Net derivative (the same family of models behind remove.bg's first release) directly in your browser via ONNX Runtime Web. The model file is about 80 MB and is cached after the first load.

U²-Net was specifically trained on portrait and product photos with fine edges, so hair and fur usually come out clean with a soft alpha. If a particular edge looks rough, lower the matting threshold in the panel — it preserves more semi-transparent pixels at the cost of a slightly noisier background.

PNG with a transparent background by default. You can also export WebP (transparent) for smaller files, or composite the cutout onto a solid colour for product shots.

No. The ONNX model runs on your device — CPU or WebGPU when available. The model file downloads from a CDN, but your image stays local. The Network tab in devtools confirms zero outbound image traffic.

Yes. You own the photos you process, and the U²-Net model is released under Apache 2.0. The cutout can be used for product listings, ads, or print without attribution.