Cutting a subject out of its background used to be a job for a steady hand and a lot of patience. Automatic background removal does it in seconds, and on the right kind of image the result is genuinely indistinguishable from careful manual work. On the wrong kind, it produces a mess. Knowing which is which saves a lot of frustration.
What the software is actually doing
It is not detecting "background" in any literal sense. It has been trained on an enormous number of images with the subject already marked out, and it predicts, pixel by pixel, how likely each one is to belong to the subject rather than the surroundings. That prediction is excellent where its training data was rich and where the boundary is obvious.
This explains both its strengths and its failures. A person standing against a wall is the most common photograph in existence, so models handle it superbly. A pane of frosted glass in front of a similarly coloured wall is rare and genuinely ambiguous, so they struggle.
What works well
- People and portraits. Overwhelmingly the best-handled case.
- Products on a plain surface. Exactly what the technique was built for.
- Animals. Usually good, with the caveat about fur below.
- Anything with a clear colour or brightness difference between subject and surroundings.
What causes trouble
Fine hair and fur
The classic hard case. Individual strands are thinner than a pixel in places, and each of those pixels is genuinely part hair and part background. Modern models handle this far better than they used to, but wispy flyaway hair against a busy background will still produce either a halo of leftover background or a slightly chewed outline.
Transparent and reflective things
Glass, water, cellophane, sunglasses. The pixels behind them genuinely are background, seen through the subject. There is no correct answer for what should happen, so the software picks one and it often looks wrong.
Low contrast between subject and background
A grey jacket against a grey wall gives the model very little to work with. If the edge is hard for you to see, it will be hard for the software too.
Motion blur and soft focus
A blurred edge has no definite boundary to find, so the cut tends to land somewhere arbitrary within the blur.
Getting a better result
- Start with the highest-resolution version you have. More pixels along the boundary means a more precise edge. Do the cut-out first and shrink afterwards, never the other way round.
- Shoot against contrast where you can. If you are taking the photo yourself, put the subject against something clearly different in colour and brightness. Five seconds of thought at capture beats ten minutes of repair later.
- Avoid a background that matches the subject. A white shirt on a white wall is asking for trouble.
- Keep the edge in focus. Sharp boundaries cut cleanly; soft ones do not.
Save it as a PNG, not a JPG
This trips up more people than the cut-out itself. JPG cannot store transparency. There is no alpha channel in the format at all. Save your cut-out as JPG and the transparent area is filled in with solid white, which is exactly the white box people complain about when they place a logo on a coloured background.
Save as PNG, which supports full transparency, and the subject will sit on any background you put it on. WebP also supports transparency and produces smaller files, if the destination accepts it.
Removing a background versus making a colour transparent
These are two different jobs and the right tool depends on which you have.
Background removal identifies the subject and cuts everything else away. Use it for photographs, where the background is complex.
Making a colour transparent knocks out pixels matching a colour you pick, with a tolerance setting controlling how close a match has to be. It is the better choice for logos, diagrams and scanned line art on a flat white background — the edges stay perfectly crisp because nothing is being predicted. Raise the tolerance if a fringe of off-white survives; lower it if parts of the subject start disappearing.
Common questions
Why does my cut-out have a faint outline of the old background?
Colour fringing. Pixels right on the boundary were a blend of subject and background, so a trace of the old colour survives. It is most visible when you place the cut-out on a background of a very different colour. A higher-resolution original reduces it considerably.
Can I put the subject on a different background?
Yes — the Remove Background tool includes a "use a photo" option so you can swap in a new backdrop directly rather than doing the cut-out and composite in two steps.
Why did it cut off part of my subject?
Usually low contrast at that point, or something the model did not recognise as belonging to the subject — an outstretched arm, a held object, a thin strap. Backgrounds with more contrast give consistently better results.
Is it better than doing it by hand?
For most photographs, it is faster and the result is comparable. For glass, smoke, or fine hair against a busy background, careful manual work still wins. The honest answer is that automatic removal handles the ninety per cent of cases that are not interesting, which is exactly what you want from it.