Remove backgrounds.

Self-hosted BiRefNet. Your images never leave your server.

No upload No sign-up No watermark Runs on your server

Drop an image to remove its background

click to choose · · paste (Ctrl/⌘ V) · or try a sample

Drop many files, a folder, or a .zip → bulk grid with per-image edit & ZIP export

Export for

Background

Gradient templates

Studio scenes

Add contact shadow

Refine mask (brush)

Paint over the cutout to fix mistakes — keep area the model removed, or erase area it left behind.

Wand tolerance 30 Contiguous (connected pixels only) Brush size 30 px Hardness 80% Opacity 100%

Drop shadow

Enable drop shadow X offset 0 px Y offset 8 px Blur 16 px Opacity 35%

Edge refinement

Feather 0 px Erode/grow 0 px Threshold 128 Hard cutoff (sharper, no anti-alias) Decontaminate edge colours Decontam strength 85%

Output

Preset Format Quality 95% Crop to subject bbox Crop padding 20 px

How it works & tips

BiRefNet (Bilateral Reference Network) is currently the best open-source background-removal model. It produces clean mattes for hair, fur, and translucent edges.
Everything runs locally in your browser. Your images never get uploaded.
First run downloads the model (~180 MB for BiRefNet-lite, 880 MB for the full model). It's cached afterwards.
Use WebGPU for 3–10× faster inference. Chrome, Edge, Safari, and Firefox now support it.
For best results, use high-resolution input. The model resizes internally to 1024×1024 (lite) or 2048×2048 (HR) and upscales the mask back to original.
Hair/fur edges are usually clean. If you see haloing, increase Erode by 1–2 px or enable a hard threshold.

Batch