Map Extract
Free tool

Remove the background from a photo

Choose a photo and press one button. A segmentation model running in your browser finds the subject, removes everything behind it and lets you save the result with a transparent, white or coloured background.

Runs in your browser. On first use it downloads an open-source processing engine from a public code CDN; your files stay on your device.

Removing a background

  1. Choose a photo

    Any picture up to 6,000 pixels on a side. Its size and file size are shown once it opens.

  2. Pick a model

    Fast (U2-Netp) is a 15.2 MB first download in total and quick on any device, with softer edges on hair. Detailed (Silueta) is 52.9 MB in total and gives cleaner edges, but is slower on phones.

  3. Run it

    Press Remove the background. A progress bar shows the one-off download, then the cut-out appears beside the original on a chequered pattern.

  4. Tidy and save

    Raise Edge clean-up to remove faint halos, or set it to 0 to keep the model's soft edges. Choose Transparent, White or Colour, then Download PNG.

Example: product shots for a stockist proposal

A small candle maker is writing to independent gift shops about stocking her range. Her photos were taken on a kitchen table, and she wants the products on plain white so they sit neatly in a one-page PDF.

On a laptop she chooses the Detailed model for cleaner edges around the wicks, waits for the first download, then sets the background to White. The file saves as the original name with -no-background added, as a PNG at the photo's full size. The second and later photos skip the download, because the browser has kept the model.

Input
candle-amber.jpg, Detailed (Silueta), edge clean-up 10, background White
Result
candle-amber-no-background.png, same pixel size as the original

What the model is good and bad at

The model looks at a small square version of the photo to decide which pixels belong to the main subject, and that mask is stretched back to full size. It does best with one clear subject against a contrasting background. Fine hair, glass, smoke and objects that blend into the background are harder.

Under the result, a note gives the share of the picture the subject covers. A very small share usually means the model could not find a clear subject; try a tighter crop, or the other model. Check edges at full zoom before you use a cut-out in print.

Questions people ask

Is my photo uploaded to remove the background?

No. The model runs inside your browser, on your own processor. What is downloaded is the open-source ONNX runtime and the model weights, from the jsDelivr public code library, on the first run. Your photo stays on your device the whole time.

Why is the first run slow?

On the first run your browser downloads the runtime and the model, 15.2 MB for Fast or 52.9 MB for Detailed. After that they are cached, so later photos start almost at once.

Can I put the subject on a new background?

Yes, in two ways. Choose Colour and pick any shade for a flat background, or save a transparent PNG and place it over another photo in an editor or presentation. A transparent PNG keeps soft edges, which blend better on a new background than a hard cut-out.

Where your data goes

Your photo never leaves your device. On first use the page downloads the open-source ONNX runtime and the segmentation model from cdn.jsdelivr.net, and your browser caches them for next time.

Checked against the tool on

Top