Cutting something out of a photograph is the most common editing job there is, and the one where amateur work is easiest to spot. Not because the shape is wrong, but because of the edge: a halo of the old background, a rim of color that should not be there, hair that has been shaved off in a smooth curve no head has ever had.
Getting it right is less about the tool than about knowing which of three problems you are looking at.
Three kinds of edge
Hard edges belong to manufactured things: a phone, a box, a building against a sky. The boundary is one or two pixels wide and either you are on the object or you are not. These are easy, and the mistake is treating them as if they were difficult.
Soft edges come from focus and motion. A shoulder slightly out of focus does not have a boundary at all, it has a gradient twenty pixels wide. Cut it with a hard selection and the object looks pasted on, because you have invented a sharpness the photograph never had.
Partial edges are the hard case: hair, fur, smoke, glass, a wedding veil. Individual pixels are genuinely part background and part subject. No outline can be correct, because the truthful answer is a percentage rather than a yes or no.
Most bad cut-outs are a hard-edge technique applied to a partial-edge problem.
The tools, and what each is actually for
Marquee and lasso
Rectangles, ellipses and freehand outlines. Crude, and still the right answer more often than people expect, because a huge number of jobs are “isolate roughly this area so I can adjust only it”. You do not need a precise outline to darken a corner.
Wand and similar-color selection
Click a pixel, get everything nearby within a tolerance. Excellent on flat, even backgrounds. Fragile on anything else, because tolerance is a single number applied to a picture that has gradients in it. If the background is a studio sweep it works beautifully; if it is a sky that goes from pale at the horizon to deep at the top, it will grab half of it and you will spend longer fixing the selection than drawing one.
Subject detection
Machine learning models trained to recognise what the main object in a frame is, and to return a mask for it. In the last few years these have gone from novelty to first choice for most photographs, particularly for people and animals, because they understand hair as a category rather than as a color boundary.
They fail predictably: multiple candidate subjects, reflections, glass, and anything where the photographer’s idea of the subject differs from the obvious one. Treat the result as a good first draft, not a finished mask.
Pen and path tools
Slow, precise, and unbeatable for manufactured objects with clean curves. If you are cutting out a car, a bottle or a piece of furniture for a catalogue, a path will beat every automatic method, because the correct edge is geometric and you know where it is even where the photograph is ambiguous.
Feathering, and why the default is wrong
Feather softens a selection edge by blending across a few pixels. Every tool offers it, usually at zero by default, and zero is almost never right.
Real photographs have no infinitely sharp edges. Lenses have finite resolution, sensors have finite pixels, and even a perfectly focused hard object transitions over a pixel or two. A selection with no feather at all produces the jagged staircase everyone recognises as a bad cut-out.
Practical amounts: half a pixel to one pixel for a sharp object at full resolution, two or three for something slightly soft, and considerably more for a genuinely out of focus edge. Match the feather to the blur that is already in the picture and the cut becomes invisible.
The color fringe nobody mentions
Here is the detail that separates convincing composites from obvious ones. A pixel at the boundary of a subject standing in front of a green wall is not pure subject; it contains some green. Cut a perfect outline and those contaminated pixels come with you, giving your subject a thin green rim on its new background.
The fix is decontamination: replace the color of edge pixels with a color sampled from just inside the subject, while keeping their transparency. Most serious edge refinement offers this, sometimes as a separate control. It matters most when the old and new backgrounds differ strongly in color, which is most of the time.
If a cut-out looks wrong and you cannot say why, look at the edge pixels rather than the shape. It is almost always contamination or a missing feather.
Masks, not the delete key
The single most valuable habit in this whole subject: never delete the background. Turn the selection into a mask.
A mask is a grayscale image attached to a layer, where white means visible, black means hidden and gray means partially visible. Nothing is destroyed. The pixels are still there and can come back at any point, which matters more than it sounds, because a cut-out is almost never right the first time.
It also gives you a way to work that no selection tool can match: you can paint on the mask. Painting with a soft brush at low opacity lets you bring back a wisp of hair here, hide a stray pixel there, and gradually converge on something believable. Selections are made in one shot; masks are refined over several passes.
Order of operations
A sequence that works for almost any subject:
- Get a rough selection fast, by whatever means. Do not perfect it.
- Convert it to a mask straight away, before doing anything else.
- Refine the edge: feather to match the photograph’s own softness, then decontaminate color.
- Put the subject on the new background at this point, not later. Edge problems are invisible against the old background and obvious against the new one.
- Paint on the mask to fix what you can now see.
- Check at 100 per cent and at thumbnail size. Different faults show at each.
The bit that is not technique
A cut-out fails just as often for reasons that have nothing to do with the edge. If the subject was lit from the left and the new background is lit from the right, no amount of masking will save it. If the subject was shot with a long lens and the background with a wide one, the perspective will not agree.
Before you start cutting, look at the light direction, the color temperature and roughly how compressed the perspective is. Where they disagree badly, fix that first, with a curve or a flip, or choose a different background. A technically perfect mask on a fundamentally mismatched composite still looks wrong, and viewers who could never explain why will simply not believe the picture.
