How to remove an image background without ruining the edges

Removing a background used to be careful, slow work with a selection tool. Now you upload an image and get a cutout back in a couple of seconds, and most of the time it is better than what a hurried human would produce.

Most of the time. The failures are consistent and predictable, and if you know what they are you can choose your source image accordingly rather than fighting the result.

How automatic removal decides

The software is identifying a subject and separating it from everything else. It has learned from a very large number of photographs what people, animals, products and vehicles look like, and it estimates for each pixel how likely it is to belong to the subject.

That estimation is confident where the boundary is obvious — a dark jacket against a pale wall — and uncertain where it is not. Everything that goes wrong goes wrong in those uncertain regions.

The three things that always fail

Hair is the classic case. At the edge of a head, individual strands are thinner than a pixel, so each edge pixel is genuinely part hair and part background. Handled well this produces a soft, believable edge. Handled badly it produces either a hard helmet outline or a fringe of the original background colour clinging to every strand.

Transparency is conceptually harder. A wine glass, a pair of spectacles, a bottle: you can see the background through the subject, so there is no correct answer to whether those pixels belong to it. Most tools either cut out the whole shape including what shows through, or drop the transparent parts entirely.

Low contrast is the mundane one, and it causes the most everyday disappointment. A white shirt against a white wall, a grey cat on a grey sofa. If you cannot immediately see the boundary yourself, the software probably cannot either.

• Fine hair and fur: fringing, or an unnaturally hard outline
• Glass and transparent objects: no consistent way to decide what shows through
• Subject and background of similar tone: boundary guessed and often wrong
• Motion blur at the edges: no crisp boundary exists to find
• Several overlapping people or objects: the wrong one may be chosen as the subject

Shooting for a clean cutout

If you control the photograph, ten seconds of thought beats any amount of correction afterwards.

1. Put maximum contrast between subject and background. A dark product against a light backdrop, or the reverse.
2. Keep the background plain and evenly lit, with no strong pattern and no hard shadows falling across it.
3. Separate the subject from the wall behind it so shadows land on the floor rather than the backdrop.
4. Light the subject evenly, so no part of its edge dissolves into darkness.
5. Avoid photographing a subject in front of anything the same colour as itself.

For product photography specifically, a plain sheet of paper curved up behind the item removes the visible corner where wall meets surface, and that single change eliminates most cutout problems.

Saving the result properly

A cutout only stays a cutout if you save it in a format that supports transparency. PNG and WebP do. JPG does not, and saving a transparent image as JPG replaces the transparent area with solid white, which people usually discover only after uploading it somewhere.

So the sequence is: remove the background, save as PNG or WebP, and only convert to JPG if you have deliberately placed a new background behind the subject first.

When to fix it by hand

For a marketplace listing or a social post, a slightly imperfect edge is invisible at the size it will be viewed. Spending twenty minutes refining hair that will be displayed two hundred pixels wide is wasted effort.

For a printed catalogue, a large banner, or anything placed against a strongly coloured background where fringing would show, do the manual work or reshoot. The bigger the final image, the less forgiving the edge becomes.