Convert images between PNG, JPEG, WebP and AVIF, resize, compress to a target size and strip EXIF/GPS, with ZIP download
Background Remover
Remove your images’ backgrounds in the browser with the small U²-Net model (Apache-2.0) through ONNX Runtime Web: no upload, with cutoff and softness controls, background replacement with a colour or an image, and transparent PNG download.
Workspace
How the result is calculated
The image is resized to 320×320 and prepared as in rembg: divided by its maximum value then (x − mean) ÷ std with mean [0.485, 0.456, 0.406] and std [0.229, 0.224, 0.225] in NCHW order. It is fed to the u2netp model (the small U²-Net, 4.6 MB, Apache-2.0) through ONNX Runtime Web using single-thread WASM; the first output (a 320×320 saliency map) is stretched to [0,1] with min-max and upscaled to the original size. Alpha = a linear smoothstep around the cutoff: clamp((m − (c − s/2)) ÷ s). Background: transparent or composited on a colour or an image. The engine and model (about 19 MB) load only on first use.
Limits:
Medium quality: the small model errs on hair, fur, transparent objects and busy backgrounds, and there is no manual correction brush and no larger model (RMBG is restricted to non-commercial use so I did not use it, and IS-Net/BiRefNet are large). Processing is on the CPU (WASM) without WebGPU so it takes seconds per image. The first download is about 19 MB and device memory is affected by huge images. Compositing has no advanced edge refinement. I did not test the tool on every browser (Safari/Firefox).
FAQ
Are my images uploaded?
No; the model runs in your browser.
Why is the result inaccurate?
The model is small; try adjusting cutoff and softness or a photo with a clearer background.
