How do I remove the background from a photo?

Open a photo and the tool cuts the subject out and hands back a PNG with a transparent background. The segmentation model runs in your browser, so unlike a background-removal API the picture is never uploaded — which matters for portraits and product shots you do not own the rights to publish.

Runs entirely in your browser · no upload · no account · no watermark

  1. Open a photo with a clear subject — a person, a product, an animal.
  2. The runtime and the model download once, then the cut-out is computed locally.
  3. Choose a transparent or solid background, then download the PNG.

PNG, JPEG or WebP. The picture is read by this page and never sent anywhere.

Worth knowing. The first run downloads about 19 MB (runtime plus model) and then caches it. Subject-centred photos work best; hair, fur and semi-transparent objects are the hardest cases for any model of this class.

What this page can do

When to use this page

  • A product photo or portrait with a clearly separated subject and a plain background.
  • You need one transparent PNG quickly and cannot upload the image.
  • Checking a rough cutout before doing finer work elsewhere.

When to use something else

  • Fine hair, fur, glass, smoke, transparent objects, or several overlapping subjects.
  • The subject and background are similar in colour or the edges are soft.
  • You need manual edge refinement, a mask you can edit, or a layered output.

How it compares with other options

A compact ONNX model runs in the browser here: it downloads once, is slower on large images and produces a single flat PNG with no editable mask. Photoshop’s subject selection, rembg or a hosted cutout API generally resolve hair and semi-transparent edges better and let you refine them, at the cost of installation or an upload. Expect to clean up the edges by hand for anything beyond a simple, high-contrast subject.

Where your files go

Nowhere. This page loads the code that does the work, and from then on the file you open is read by your own browser. There is no upload step, no temporary server copy and no cleanup to trust — the data never crosses the network, so there is nothing to delete afterwards. You can confirm it yourself: open your browser’s network panel while the tool runs and watch it stay quiet.

Questions about this tool

Is my photo uploaded to a server?
No. The model runs in your browser on WebAssembly. The only requests the page makes are for its own code, the 14.2 MB runtime and the 4.4 MB model — all served from this site — and none of them contains your image. Open your browser’s network panel while it runs and you will see no request at all.
What does it cost, and is there a limit?
Nothing and no. There is no API key, no per-image charge and no daily quota, because there is no server doing the work. The only limit is your device’s memory.
Which photos work best?
One clear subject that stands out from its background, filling a good part of the frame. Product shots on a plain surface and portraits with a distinct backdrop work well. Hair, fur, transparent glass, motion blur and busy backgrounds where the subject and the background are the same colour are the hard cases for any model of this size.
Why is the first run slow?
Because about 19 MB of runtime and model weights have to arrive first. The browser caches them, so the second photo on the same device starts immediately — and works offline.
Can I get a higher-resolution cut-out?
The mask is computed at 320 × 320 and then scaled up to your image’s real size, so the edges are smooth but not sharp at pixel level. For a large print, cut out at the size you need and refine the edge in an image editor — the PNG keeps a clean alpha channel to work from.
What is the model, and what licence is it under?
U-2-Netp, a small salient-object detector from the U-2-Net authors, packaged for ONNX by BritishWerewolf. The weights and the runtime are both Apache-2.0 and MIT respectively, so the output is yours to use commercially.

Other tools in the same group

Guides