Last updated 28 August 2026
How this works without a server
Background removal used to require sending your photo to someone else's machine. It no longer does. This page runs the Open Remove Background Model — an openly licensed segmentation network — through a WebAssembly runtime inside your browser, using your own processor. The image is read from your device, processed there, and written back out, never crossing the network. What does cross the network is the model file itself, fetched once from a public CDN and then cached by your browser. That is the honest caveat behind the privacy claim: your picture stays with you, but the CDN can see that someone at your address requested the model.
The trade-off is speed. The model runs on your processor rather than your graphics card — not by preference but by necessity, since this particular network uses an operation the browser's GPU backend does not yet implement. The page also avoids the cross-origin isolation headers that would allow multi-threading, because those headers break embedded third-party content. Expect several seconds on a capable desktop and longer on a phone, particularly for the first image while the model is still being fetched. Standard quality caps the image at 1500 pixels before processing and is dramatically faster if you are waiting too long.
Common questions
Is my photo uploaded anywhere?
No. The image is read, processed and saved entirely on your device. What is downloaded is the AI model file, fetched once from a public code CDN and then cached by your browser — the picture itself never leaves your machine.
Why is the first image slow?
Because the AI model has to be downloaded first — roughly 40 MB, once. On a fast connection that is a few seconds; on mobile data it can take considerably longer. Your browser then caches it, so subsequent images start processing immediately.
Why is it slower than paid tools?
Running this model at full speed needs browser isolation headers that break embedded third-party content, so this page deliberately runs it single-threaded. Paid services also run on server hardware far more powerful than a phone. The result quality is comparable; the wait is longer.
What if it fails to load?
The tool depends on a public CDN for the model, so a blocked or unavailable CDN means it cannot run. If that happens you will get a clear message rather than an endless spinner — everything else on Quezlo works without any external service.
Which images work best?
Photographs of people. The model was trained mainly on human subjects, so portraits and full-body shots come out cleanest. Products, objects and animals still work, but the edges are less reliable — especially around fine detail like hair, fur and thin straps, where every automated tool struggles.