Why does green spill turn brown instead of green on skin?
Because the bounce is not painted on top of your skin, it is multiplied by your skin and added to it. The light a green screen deposits measures 139 degrees on the hue wheel, but skin already reflects far more red than blue, so the sum lands between 40 and 91 degrees depending on how strong the bounce is: tan, khaki, brown-green. The despill slider cannot reach it either. It only acts once green overtakes red, and then it stops at red rather than at the skin's own green, which leaves 38 to 46 levels out of 255 permanently in the pixel.
The bounce is green; the pixel is not
A green screen radiating #00B140 deposits light that, once it has been filtered through skin reflectance, measures between 136 and 139 degrees on the hue wheel. Pure green is 120. So the light genuinely is green, slightly leaning toward cyan because the screen's blue channel is not zero.
But it is not deposited like paint. A surface shows you illuminance multiplied by reflectance, and the bounce is added to the illuminance before that multiplication happens. On a light-skin patch at the bounce strength we measured 0.8 m from the screen, the screen deposits 79 levels of light: plus 60 in green, plus 19 in blue, nothing in red. Skin that read (214, 171, 148) becomes (214, 231, 167). Its hue moves from 20.8 degrees to 75.7.
How brown it looks is set by how much bounce there is
Sweep the bounce on one patch of light skin and the resulting hue climbs a ladder: 0.20 gives 40.0 degrees, 0.40 gives 64.4, 0.51 gives 75.7, 0.77 gives 91.1 and 1.00 gives 95.0. Clean skin sits at 20.8. Every one of those numbers is a real mixture of green light and warm skin, not a different kind of contamination.
So the mistake passes through tan at 40 degrees, khaki at 64, and only becomes recognisably green somewhere above 90. The complaint that it looks brown-green rather than green is the middle of that ladder: moderate bounce, where plenty of green has arrived but the skin's own red is still the strongest channel in the pixel.
The despill slider often does nothing at all
Spill suppression here is a green-channel limiter, and its trigger is a single test: green greater than the larger of red and blue. On warm skin, red is already the largest channel, so unless the bounce lifts green above red, the test fails and the code takes its other branch, which is a no-op. The slider moves and the pixel does not.
The bounce needed before it fires at all, as a fraction of the key light, runs 0.33 on very light skin, 0.37 on light, 0.52 on medium, 0.60 on tan and 0.72 on deep. At the 0.51 we measured at 0.8 m, that means the control is completely inert on three of the five tones. The darker the skin, the browner the mistake looks and the less the slider can touch it.
When it does fire, it stops in the wrong place
The limiter pulls green down to max(R, B) — that is, down to the red channel — and never below. Its floor is the skin's red, not the skin's green, and those are different numbers. The residue left behind is exactly the gap between the subject's red and green reflectance times the light level: 40.8 to 45.9 levels out of 255, identical no matter how strong the bounce was.
Push spill and despill both to 100 and light skin travels from 75.7 degrees to a hard stop at 60.0 — dead yellow. It cannot go below 60, and the skin belongs at 20.8. On the numbers we measured, the sliders close 15 degrees of a 55-degree error, and then they are done.
Three channels get lit; one lever exists
On that same patch the deposit is 79 levels in total: plus 60.1 green, plus 18.8 blue, zero red. The limiter can reach 16.8 of them, which is 21.2 percent of what arrived. The blue lift of 18.8 levels has no control attached to it at all, in either direction.
Measured end to end, the default settings of spill 35 and despill 10 remove 11.6 percent of the added green. Both sliders at 100 remove 27.9 percent. After that best case, the pixel is still 47 levels away in RGB distance from what the same skin would have read with no screen in the room at all.
At close range the green channel is already gone
Heavy bounce clips before you ever reach post. On very light skin the green channel hits 255 once the bounce passes 0.62 of the key light; on light skin, at 0.71. The bounce we measured 0.4 m from a screen was 0.77, past both thresholds. The camera is no longer recording how much green there is, only that there is at least this much.
Once green is clipped, the information is gone. The limiter can still pull 255 down to the red channel, which lands you on exactly the residue described above, but no slider setting recovers the lost headroom. If you want to know whether this happened, look at the green histogram of the subject rather than at the image.
What this means for the shot
The spill control is genuinely worth using: on light skin at 0.8 m it takes the total error from 56 levels down to 47. It is a partial fix on a problem that was created at capture time, and the part it cannot fix is structural — it is the distance between the subject's red reflectance and its green reflectance.
The levers that actually work are physical ones. Move the subject back, flag the screen off the parts of the body that can see it, or turn the screen's own lighting down. Our distance test put the screen's share of the light falling on a side-facing surface at 57 percent at 0.4 m and 24 percent at 4 m. And there is a quick check: sample a patch of skin and look at green against red. If green is below red, the slider will do nothing, and the fix is upstream.
Related questions
Does spill suppression remove green spill completely?
No. Measured on light skin at the bounce strength we measured 0.8 m from the screen, the default settings of spill 35 and despill 10 remove 11.6 percent of the added green. Both sliders at 100 remove 27.9 percent. The remainder sits below the red channel, where a green-channel limiter cannot see it.
Why does the despill slider do nothing on darker skin?
Because it only fires when green is greater than the larger of red and blue. Warm skin has red as its strongest channel, so the bounce has to lift green above red first. That takes a bounce of 0.33 on very light skin but 0.72 on deep skin. At the 0.51 we measured at 0.8 m, medium, tan and deep skin are all below the trigger.
Is a brown ring a keying problem or a lighting problem?
Lighting. Across every skin tone and bounce strength we modelled, the closest any contaminated pixel came to the key colour was 0.487 in normalised RGB distance, while the key only starts to soften pixels at 0.259. Nothing was at risk of being keyed out. The colour was wrong; the alpha was fine.
Should I just push spill to 100 and leave it there?
It helps, but it has a hard floor. Both sliders at 100 take light skin from 75.7 degrees to a stop at 60.0 degrees, which is pure yellow, and leave 43 levels of green in the pixel. If the patch already looks brown rather than green, the slider will do nothing at all, because green is still below red.