How it works
- Add photos. The first time, the AI model (about 140 MB) is downloaded to this browser and kept there; a progress bar shows how far it is. Next time it starts straight away.
- Wait a few seconds per photo. In our test on an Apple M2 Mac in Chrome, photos from 512 × 512 to 4,000 × 3,000 took 3.6–7.8 seconds each; slower graphics chips take longer. The time is shown on each result.
- Pick the background and format — transparent, white or any colour; PNG, WebP or JPG. Changing them redraws the results instantly, without running the AI again.
- Download each photo, or all of them as a ZIP.
What gives the cleanest result
- One clear subject that stands out from the background: a person, a product on a table, a pet.
- Good light and a sharp subject. Very fine hair, glass and smoke are the hardest cases for any background remover.
- Close-ups where the whole frame is the subject (a texture, the body of an instrument) have no background to remove; the result can look odd.
Why it runs on your device
Most background removers upload your photo to a server, and limit free use with sign-ups, daily caps or small previews. Kilomill runs an open-source model, BiRefNet (MIT licence), in your browser with WebGPU. Your photos stay on your device, results are full resolution, and there is no limit on how many you process. The price is the one-time download and a browser with WebGPU: current Chrome and Edge, and Safari 26 or later. Phones and older computers may be slow or run out of memory.
After removing the background
Need a product photo at an exact size for a marketplace? Download the PNG, then resize it or get it under a size limit. For a website favicon or logo, convert the PNG to ICO.
Updated