AI Background Remover

Yes — an actual alpha channel. If you export to JPEG afterwards the transparency becomes white, which is a property of JPEG rather than of the cutout.

That is what the fox and character examples above are there to show. Partial-coverage edges are kept rather than clipped, which is the difference between a cutout and a sticker.

Transparent subjects are the hardest case, and the perfume bottle above is the example. The bottle stays see-through while the water and smoke behind it go.

The same credit ladder as everything else — the estimate shows before you run it, and a failed run is refunded.

Yes, and often you should skip the cutout entirely: the image editor can replace a background in one sentence without an intermediate PNG.

It will, but the edge quality follows the input. If the source is small, run the upscaler first.

Almost always because of partial pixels. At the boundary of a subject, a single pixel is part subject and part background — a hair strand might cover a third of one. A hard cutout has to decide all-or-nothing for each of those, and the accumulated decisions read as a cut-paper edge. Keeping the partial value is what makes an edge look like it belongs to the picture.

As high as you have. The edge is computed from the pixels you provide, so a small source gives a coarse boundary no matter how good the model is. If your only copy is small, run it through the upscaler first and cut out the enlarged version.

It separates foreground from background, not one foreground object from another. If two people overlap, they come out together. For that kind of split, describing the change in the editor usually works better than a cutout.

A cast shadow belongs to the background and normally goes with it. A shadow on the subject stays, because it is part of how the subject is lit — and removing it would make the cutout look flat when you place it somewhere new.

Yes, and usually better than on photographs — flat graphics have hard boundaries, which is the easy case. The difficulty scales with how soft the edge is, so a logo is trivial, a person is moderate, and a long-haired subject against a busy background is the hard end of it.

One at a time through the browser. For a full catalogue the practical approach is a session rather than a single action, and the cost is per picture — the estimate before each run is the same number every time, so the total is easy to predict before you start.