Guide
Chroma Key: Removing a Background
Short answer: a chroma key removes a background by asking the same question about every pixel in the image — how far is this pixel’s colour from my key colour, ignoring how bright it is? Nothing else is involved in the decision. Pixels inside a radius around the key become fully transparent; pixels outside a wider radius stay fully opaque; pixels between the two get a partial value. Tolerance sets that inner radius, edge softness sets the gap to the outer one, and spill removal is a separate clean-up that happens after the decision has already been made. Once you see the removal as that single question, every slider in every tool stops being a mystery — and the failures stop being mysterious too.
The part that surprises people: brightness is deliberately ignored
The question is asked in colour only. Each pixel is reduced to two colour values that describe its hue and saturation, with the brightness component set aside before the distance is measured. That single design decision is what makes the tool usable on a real photograph, because a real green screen is never evenly lit — there is a bright patch where the light falls and a darker patch in the corner, and those two areas are the same colour at two different brightnesses.
If brightness were included in the measurement, the bright patch and the dark patch would sit at different distances from your key colour, and one tolerance setting could not delete both. Ignoring brightness collapses them onto the same point, which is why one setting can clear a screen that looks visibly uneven to your eye. It is also why a dark green object and a lit green screen can end up at the same distance from the key — the mechanism genuinely cannot tell them apart, and no amount of slider adjustment will teach it.
What each control is doing, in terms of that one question
| Control | What it changes about the question | What you see when it is wrong |
|---|---|---|
| Key colour | Where the centre of the measurement sits | Screen survives everywhere, or the subject is being eaten from the first click |
| Tolerance | The inner radius — how far counts as background | Too small: patches of screen left. Too large: holes start appearing in the subject |
| Edge softness | The gap between the inner and outer radius | Too small: a stair-stepped, cut-out-looking outline. Too large: a blurry halo that no background hides |
| Spill removal | Nothing — it runs after the decision | Without it, a green rim survives on hair and shoulders even though the key itself is correct |
The last row matters most, because it is the one people try to solve with the wrong control. Spill is not a leftover of the background; it is green light that physically landed on the subject. Those pixels are genuinely part of the person, so the key is right to keep them. They need to be de-coloured, not deleted, which is a different operation and therefore a different slider.
Three consequences you can predict from the mechanism
Green screens are usually easier than blue ones, and it is not superstition. Every pixel is stored as three colour channels, and in the way those channels are weighted to reconstruct brightness, the green channel carries the largest share. More of the signal survives for the other two channels to describe colour with, so the boundary between subject and screen tends to arrive with more usable information. The exception is a subject that is itself green — then the advantage inverts, and blue is the better screen.
How far you move the camera changes the edge, not the screen. The further the subject stands from the wall, the more the screen sits in the same focus plane as the foreground, and the crisper the boundary that reaches the key. A subject pressed against the screen blurs into it, and a blurred boundary is the one case where a wider softness band is genuinely the right answer rather than a compromise.
A shadow on the screen is not one problem, it is two. The shadowed area is the same colour at a lower brightness, so it is handled by the brightness-independence described above. But where the shadow has a soft edge, the transition band becomes wide, and the same softness setting that cleans the subject’s edge will now be applied to a much wider zone. This is why some frames need the screen re-lit rather than the settings re-tuned.
Where this mechanism stops working
Because the question is about colour alone, the mechanism has no concept of a subject. It cannot tell a green screen from a green shirt, a green plant from a green wall, or a carefully lit screen from a painted one. Everything the tool does well follows from that simplicity, and everything it cannot do follows from it too — there is no setting being withheld, and no version of the algorithm that would behave differently. The frames where this breaks down are covered here, including the four failure patterns and the checks that catch them before you spend time on a cut-out that cannot be saved.
What replaces the missing judgement is the two-background check: put the result on black to see holes, and on white to see halos. That test is not a workaround for a weak tool; it is the human half of the operation, and it catches exactly the two mistakes the per-pixel question cannot warn you about.
Questions
What is the difference between chroma key and colour key?
They describe the same mechanism: declaring one colour to be background and deleting pixels that are close enough to it. Different programs use different names for the same per-pixel distance measurement, which is why a setting called tolerance in one place is called something else somewhere else. Match on behaviour, not on the label.
Why does my key remove part of the subject as well?
Because those pixels really are close in colour to the screen, and the mechanism only measures colour distance. A green logo, a green plant, an olive-toned jacket and a shadowed area of clothing can all land inside the radius. The fix is either to narrow the radius and accept some leftover screen, or to change the key colour if your subject allows it.
Does a higher tolerance give a cleaner result?
No. Tolerance moves the whole decision outward, including the part that was already correct. Raise it until the screen is fully gone, then stop and solve whatever is left with the edge and spill controls, which affect different pixels.
Why is the edge still visible if the background is fully removed?
Two different reasons, and they need different controls. A hard, stair-stepped outline means the transition band is too narrow for the resolution of the image. A pale coloured rim means spill. Looking at the result on a light background tells you instantly which of the two you have.
Can chroma key remove a background from a photo that has no screen?
No, and the failure is structural rather than a matter of settings. With no screen there is no colour to nominate, so nominating one anyway means nominating part of your subject. Removing an arbitrary background is a recognition problem, which is a different operation with different failure modes.
Open the green screen remover — the four controls described here, running in your browser, exporting a full-resolution transparent PNG. If you have not picked the key colour yet, start with choosing it; if you are deciding between this route and a full editor, the comparison is here.