What's the actual difference between NearBG and remove.bg?

The difference is architectural. remove.bg is a cloud service: the model runs on their servers, so every photo has to be uploaded before it can be processed. NearBG turns that around — your browser downloads a small segmentation model (U2Netp, about 4 MB) once and runs it on your own device via WebAssembly, so the photo itself never travels anywhere. The result is the same kind of cutout — a transparent PNG, plus a manual touch-up brush for stray edges — with no upload, no account, and no server in the loop.

The same job, step by step — without the upload

1. With a cloud remover like remove.bg, step one is always the upload: your photo travels to their servers before anything happens. With NearBG, step one is the opposite — the model travels to you, as a one-time download of about 4 MB from our own domain.
2. remove.bg then runs its model on their hardware; NearBG runs U2Netp on yours, via WebAssembly in your browser. Same class of task — segmentation — different place for it to happen.
3. A cloud service has to send the result back to you and decide how long to keep your original; NearBG has nothing to send back and nothing to keep, because your photo never left the tab.
4. Both give you a cutout on a transparent background. NearBG adds a manual touch-up brush — and never adds a watermark or a per-image charge.

The honest trade-offs of on-device versus cloud

Doing everything in the browser is a real architectural constraint, and it cuts both ways. The upside: nothing is uploaded, so there is nothing to intercept, retain, or leak — the page's Content-Security-Policy technically blocks image data from leaving, which means privacy is enforced by your browser, not merely promised in a policy document. It also means no account, no quota and no watermark, because there is no per-image server cost to recover.

The downside, stated plainly: a cloud service can throw a far larger model at your photo than a browser can comfortably run. NearBG's U2Netp is compact (about 4 MB) and computes its mask at 320×320, so on very fine edges — stray hairs, fur, translucent fabric — a large cloud model may produce a cleaner automatic edge. NearBG's answer is the touch-up brush plus zero cost per attempt: you can re-run and hand-fix as often as you like without burning credits. Which trade is right depends on the photo — and on whether that photo is something you'd rather not hand to someone else's server.