Why is the chroma key green #00B140 and not just any green?
Because two different things are voting, and they want opposite ends. The camera wants the screen bright, so the sensor has something clean to record: #00B140 sits at luma 131, roughly double the 64 of a blue screen. The keyer wants the screen far away from your skin and clothes: #00B140 keeps 0.669 of normalised distance to the nearest skin tone, against a key window that only reaches 0.259. Pick pure green #00FF00 instead and you get more distance but no headroom at all — its green channel is already pinned at 255, and in our test scene 76.85% of the background in the falloff corners survived the key.
What the keyer actually measures
Before any colour can be defended, the measurement has to be pinned down. The keyer in this tool compares each pixel to the key colour as a point in normalised RGB space — every channel divided by 255, then the straight-line distance between the pixel and the key colour. That is the entire decision. Pixels inside the radius disappear, pixels outside stay, and pixels in between fade out gradually.
Two sliders set that radius. Similarity 42 maps to a tolerance of 0.210, and Smoothness 14 widens the fade by 0.049, so the working window is 0.161 to 0.259 in the same units. Everything below is measured against that window: a pixel closer than 0.161 to the key colour is fully removed, a pixel further than 0.259 is fully kept.
That window is narrow, and it is the reason the colour matters. Any colour you pick has to survive two separate attacks: the lighting on the screen is never perfectly even, and your subject is never infinitely far away in colour space.
Why the camera gets a vote
Rec.709 weights luminance at 0.2126 red, 0.7152 green and 0.0722 blue. Green carries about 71.5% of the brightness of anything a camera records. A green screen is therefore the brightest screen you can light with a given lamp, and brightness is what keeps sensor noise from swallowing the fine detail at the edge of your subject.
Measured on that scale: #00B140 lands at luma 131.2. The blue screen preset #0047BB lands at 64.3 — less than half. Under the same lamp and the same exposure, the green screen gives the sensor roughly twice the signal, which is the same as saying half the relative noise. It is also why a Bayer sensor helps green along: in the standard RGGB pattern, half the photosites are green, so the green channel is recorded at twice the sampling density of red or blue. That last point is a published property of the colour filter array, not something measured here.
Brightness alone would argue for the brightest green available, and that is exactly where it breaks. #00B140 has a green channel of 177, which leaves 78 levels of headroom before that channel saturates at 255. In our run the colour only fell out of the key window once the screen was lit 1.355× past reference. Push #00FF00 instead and its green channel starts at 255: there is nothing left to record, so any part of the screen brighter than reference clips to the identical value. The screen stops telling you anything about how it was lit.
Why pure green is the worst green
This is the part that surprises people, because pure green looks like the obvious choice — it is the furthest thing from skin in the whole sRGB cube, at 0.878. Distance is not the binding constraint. Evenness of light is.
We rendered a 640×360 test scene: a subject with a feathered 2-pixel edge and three thin hair strands, in front of each candidate screen, with lighting falling off 30% from centre to corner and sensor noise of σ=3/255 added. Then we ran the real keyer over it at the default settings and counted background pixels that survived.
#00FF00 left 52.07% of the background unremoved, and 76.85% in the falloff corners. #00B140 left 22.95% overall and 37.76% in the corners — less than half the damage. A deeper #008000 left 5.93% and 9.77%. The pattern is consistent: the brighter and more saturated the screen, the more a dim corner drifts out of the key window.
Push the falloff to 40%, the ordinary result of a small room with one lamp, and pure green gives up entirely: 66.59% of the background survives, 93.41% in the corners. #00B140 gives 51.08% and 76.19%. In arithmetical terms, #00FF00 loses the screen once the light drops to 74.1% of reference, while #00B140 holds to 64.9% and #008000 holds all the way down to 48.4%.
So why not just use the dark green
Because the dark greens win the keying argument and lose the photography argument. #008000 was the most robust screen we tested — 5.93% background survival, 1.520× overexposure tolerance, and it holds the key down to 48.4% lighting. It is the colour you would pick if lighting were free and cameras were noiseless.
Lighting is not free. At luma 91.5 it is 30% darker than #00B140, so every imperfection in the sensor is relatively larger, and it also sits closer to skin: 0.595 to the nearest skin tone against 0.669 for #00B140. You would be trading a cleaner key on an evenly lit studio wall for a worse one on a real subject in a real room.
#00B140 is not the winner of any single test in this article. It is the colour that loses the least across all of them.
Where the compromise actually bites
The sharpest limit is not the screen at all — it is the half-covered pixels along your subject. Take a pixel that is 50% subject and 50% screen and measure how far it lands from the key colour. If that distance is under 0.259, the keyer will start removing a pixel that is genuinely half your subject.
For #00B140 and fair skin that distance is 0.549, comfortably safe. For #00B140 and blonde hair it is 0.471, still safe. For #00B140 and denim blue #3B5C7E it is 0.237 — under the 0.259 line. Jeans are the one common garment that quietly fails here: the soft edge of a denim leg gets eaten before you notice.
The same blend against #008000 measures 0.282, which clears the line, and against #00FF00 it measures 0.420, the safest of the greens. It is a genuine trade: the brighter greens give you more separation from blue clothing and less tolerance for uneven light.
When blue is still the better screen
Blonde hair is the classic case, and the numbers agree with the old craft rule. Measured from the same colour, blonde sits 0.942 away from #00B140 but 1.070 away from #0047BB. Blue screens also survive the falloff test slightly better than the preset green — 19.13% background survival against 22.95% — because their blue channel is not carrying the whole signal.
The costs are real though. Blue sits at luma 64.3, so you are buying that separation with half the light and you will need more of it. And blue clothing becomes the problem instead of green: denim measures 0.343 from #0047BB, and the half-covered blend at a denim edge falls to 0.171, far below the 0.259 line. Blue eyes go the same way. If your subject is in jeans, the blue screen that saved the hair will cost you the legs.
Can you key against any colour at all
Yes, and that is not a figure of speech. As one long-standing answer on Photography Stack Exchange puts it, "green" screen is a generic term now, the original travelling mattes were more blue, and you can use most any colour instead — what matters is that nothing on your subject shares it.
Our runs confirm the mechanism at both extremes. A black screen has no brightness to lose, so uneven lighting cannot move it: 0.00% background survived, even at 40% falloff. A white screen is the opposite, and the worst screen we measured: 76.38% survival at 30% falloff and 100% in the corners, because every shadow on it is a colour change.
Both are available as presets here, and both need a tighter Similarity than the coloured screens. The reason they are not the defaults is the same reason a mid-bright green is: a flat black or white backdrop is rarer in practice, and any wrinkle in it reads as a completely different colour.
What this means for a real shoot
The colour matters less than you think and the lighting matters more than you think. Every screen in this article failed the same way — not because the key colour was wrong, but because a corner of it was darker than the rest. Moving your light 30% further away costs more than picking the wrong green.
If you are choosing, choose for the subject. Green for skin and dark clothing, blue for blonde hair and green clothing, and check denim either way. If you are shooting jeans against green, drop Similarity a few points rather than moving the lamp: at 42 the denim edge is already inside the fade, and the fade is what removes it.
And leave the headroom alone. The 78 levels between 177 and 255 are the reason an overexposed patch is still recognisable as screen rather than a blown-out white hole. Any preset that starts at 255 has already spent it.
Related questions
Is #00B140 an official broadcast standard?
We have not found a first-party source that names it as one, so we do not claim it. What we can show is why a colour in this range behaves well: luma 131, a green channel of 177 with headroom left, and clear separation from skin tones. It is the preset in this tool because of those measurements, not because of a standard number we can cite.
Why does my green screen key perfectly in the middle and badly at the edges?
Light falloff. In our test scene with 30% falloff, #00B140 left 22.95% of the background behind overall but 37.76% in the corners, and at 40% falloff it was 76.19% in the corners. The key colour is the same everywhere; the recorded colour is not.
Should I switch to a blue screen for blonde hair?
The numbers support it: blonde sits 1.070 from #0047BB versus 0.942 from #00B140, and the blue screen survived falloff slightly better. Check what the subject is wearing first — denim drops to 0.343 from blue, and its soft edge to 0.171, which is well inside the key window.
Can I use a black or white background as a key?
Both work mechanically. Black measured 0.00% background survival in our falloff tests because it has no brightness to lose; white was the worst screen measured, at 76.38% overall and 100% in the corners. Both need a tighter Similarity than a coloured screen, and both punish wrinkles and shadows hard.
Does raising Similarity fix an uneven screen?
It hides it and charges you for it. Widening the radius pulls the dim corners in, but it also pulls in any pixel that is genuinely part of your subject — at the default 42, a half-covered denim edge already sits at 0.237, inside the 0.259 line. Light the screen evenly instead.