Background remover

How to use it

1. Choose a photo with a clear subject.
2. Wait for the cut-out. The first time, your browser downloads the model (about 4.5 MB) and keeps it.
3. Check the edges against the checkerboard preview.
4. Download the transparent PNG.

Your photo never leaves this tab

This is not a wrapper around somebody’s API. The segmentation model is downloaded to your browser and run there, on your own machine, on your own hardware — the photo is never uploaded to Sellstamp or to anyone else, because there is nothing here to upload it to. The only thing that travels is the model, and it travels towards you.

The first run downloads the model — 4.5 MB, once — and your browser keeps it afterwards. On a browser with WebGPU it then takes about a second per photo; on one without, a few seconds on the CPU. There is no queue, no credit, and no limit on how many photos you put through it, because none of them cost us anything.

What it is good at, and what it is not

The model is U²-Net, a general salient-object model rather than a portrait-only one, so it handles product shots, people, food, furniture and animals. It is strongest with a clear subject against a distinguishable background.

Its one structural limit is resolution: it looks at your photo at 320 pixels square and the resulting outline is scaled back up, so edges are softer than a much larger model would give. On a product against a plain background you will not notice. On fine hair against a busy background you will.

It struggles where every matting model struggles: that fine hair, transparent and reflective objects like glass and cellophane, motion blur, and a subject the same colour as what is behind it. For a catalogue, a plain sweep behind the product will give you a cleaner cut-out than any amount of post-processing.

What you get back

A PNG with a real alpha channel, which is what “transparent” means — so it composites onto any background later. JPEG cannot store transparency at all; saving a cut-out as a JPEG puts a white box straight back behind it. Very large photos are reduced to 4096 pixels on the long edge, which is far beyond any web use of a cut-out and keeps the tab responsive.

Selling the product in the photo? Sellstamp’s product pages want exactly this: a cut-out on a transparent background, which sits cleanly on a storefront in either a light or dark theme.