What removal is actually solving
For every pixel, a removal tool decides how much of it belongs to the subject. For most pixels this is easy — deep inside the subject or well outside it. The difficulty lives entirely in the transition band at the edge, which is rarely a clean line.
A single pixel at the boundary captured light from both the subject and the background, so its value is a mixture. Estimating the true mixing proportion, and the subject's underlying colour without background contamination, is the hard part of the problem. It is called matting, and it is much harder than segmentation.
Everything below follows from one principle: make the transition band as short and as unambiguous as possible.
Contrast is the whole thing
The single highest-leverage decision is the difference between subject and background — in brightness first, then in colour.
A dark subject against a light background separates trivially. A dark grey subject against a mid-grey background does not, and no tool will do it well, because the information required is genuinely not in the image. If you photograph a black jacket against a black sofa, you have not made a hard problem for the software; you have made an unsolvable one.
Brightness contrast matters more than hue contrast, because edge detection operates primarily on luminance. Two colours that look very different but have similar brightness — a mid red against a mid green — separate worse than you would expect.
The practical move is simply to look at the background before shooting and ask whether the subject's edges are visible against it. If you have to look hard, so will the algorithm.
Lighting the edges
The second lever is where the light is, and the goal is to keep the subject's outline bright and distinct rather than falling into shadow.
Front-lit subjects separate best. Light coming from behind the camera means the edges facing you are lit, so the outline is well defined against the background.
Backlighting is the hardest case: the subject becomes a silhouette with a rim of flare, edges lose definition, and the transition band grows wide and ambiguous. Backlit photos look dramatic and cut out badly.
A rim light — a light behind and to one side, catching the subject's outline — is the professional trick for exactly this problem, and it is why studio cut-outs look effortless. It draws a bright line around the subject that separates it from the background regardless of tone.
Also, keep the subject away from the background so the background falls out of focus slightly. Blur at the boundary is a real cue: the sharp subject and the soft background are easy to distinguish, and shadows cast onto a distant background do not appear in the frame.
The materials that are genuinely hard
Some subjects are difficult for every tool, and knowing which lets you plan around them.
Hair and fur are the classic case. Individual strands are thinner than a pixel, so a large fraction of the outline is semi-transparent — the transition band is not a thin line but a wide, fuzzy region. This is why hair is the benchmark everyone uses. Shooting hair against a strongly contrasting, evenly lit background is the only real answer.
Transparent and translucent materials — glass, bottles, plastic, sheer fabric, ice — break the underlying assumption that each pixel is either subject or background. A glass is both: you see it and you see through it. No automatic tool handles this well, and the practical approach is to photograph glassware on the background you actually want, or to accept manual work.
Motion blur widens the transition band across the whole moving region rather than just the edge. Reflective surfaces such as chrome show the background in the subject, which confuses tools in a way that is hard to correct. Fine repeating structures like wicker, mesh, lace and chain link have enormous edge length relative to their area, so small per-edge errors accumulate visibly.
A shooting checklist
In order of impact:
- Pick a background that contrasts strongly in brightness with the subject.
- Use a plain, untextured background — a wall, a sheet, a roll of paper.
- Avoid backgrounds containing colours that appear in the subject.
- Light from the front, not from behind. Add a rim light if you can.
- Move the subject well away from the background to soften it and to move shadows out of frame.
- Use even, soft lighting so no part of the outline falls into deep shadow.
- Shoot at the highest resolution you have — more pixels across the transition band means a better matte.
- Keep the subject in focus throughout; a soft edge is an ambiguous edge.
- For hair, avoid backgrounds close in tone to the hair colour, which is the single most common cause of a poor cut-out.
Checking the result properly
Most cut-out flaws are invisible in the one view people check them in.
A subject cut from a white background, previewed on white, will look perfect even when it has a significant white fringe — the fringe is the same colour as what you are viewing it against. Always preview against a background very different from the original, and ideally against both black and white.
Zoom to 100% and inspect the edges, particularly around hair, shoulders and any fine detail. At fit-to-screen the errors that matter are all sub-pixel.
Look specifically for three things: a halo of the original background colour, a hard jagged edge where the transition band was collapsed to binary, and detached fragments where the tool lost a thin element like a strap or an antenna. All three are fixable with a refine brush if you catch them; none are fixable after you have published.