Which model removes the background
The tool cuts you out of the photo and puts white behind you. That is done by a neural network running inside your browser, and you choose which one.
Why there is a choice at all
A bigger model traces hair, glasses and thin edges more carefully, and costs more to download. The light one runs first because 5 MB is a fair price for a result most photos are fine with. If your outline came out ragged, switching model is the fix, and the tool offers it right after the first attempt.
| Model | Download | Best for | On this device | |
|---|---|---|---|---|
| U2Net-PDefault | 5 MB | Plain backgrounds, short hair. The default. | … | |
| Silueta | 43 MB | Same idea, a little more careful at the edges. | … | |
| U2Net Full | 176 MB | Busy backgrounds where the light model bleeds. | … | |
| U2Net Human | 176 MB | Trained on people: the best one for hair and glasses. | … | |
| ISNet General | 176 MB | Fine detail, at 1024 px input instead of 320. | … |
The default is what the background button reaches for first. Downloading a model here, on wifi, means it is ready when you need it instead of arriving mid-job.
Where the model is kept
It downloads once and stays in your browser's storage, so the second photo needs no network at all. Clearing site data removes it. The photo itself is never uploaded — only the model comes down.
What no model fixes
A patterned wall, a shadow behind your head, or clothing the colour of the wall will defeat every model here. Standing a metre away from a plain wall beats any amount of megabytes.
This is an independent site, not a government body. Requirements change, so check them against the official source before you apply. We promise a photo that matches the published specification. We never promise that an application will be approved.