Why does a shiny subject mirror the whole green screen back at the camera?
Because a curved mirror folds the whole room onto itself, and a cyclorama is mostly screen. On a chrome sphere standing inside one, the mirror direction is already 35 degrees off the lens axis at three tenths of the radius and 60 degrees at half, so every part of it except the patch facing straight back at the camera shows the screen: an average of 34.1 code levels of green above its own red and blue, peaking at 44.5, and 52.6 with spill suppression off. It rarely punches holes at the default settings — it makes the object the wrong colour. Open similarity to 100 to chase the screen and 39.8 per cent of the object goes with it, with damage reaching seven tenths of the way in from the silhouette.
A curved mirror folds the whole room onto one surface
Every point of a glossy surface shows you the scene from a different angle, along the mirror direction, which is the view direction bounced off the local normal. On a sphere that swings fast: the direction is 35 degrees off the lens axis at three tenths of the radius, 60 degrees at half, and 89 degrees at seven tenths. A flat object sends one reflection back; a sphere sends you the entire surroundings, compressed.
If those surroundings are a cyclorama, most of what it sends back is screen. Measured on a chrome sphere inside a wraparound set that leaves only a thirty degree opening for the camera, the green it carries runs minus 2.0 levels inside a fifteenth of the radius, 10.4 by three tenths, and then 34.6, 35.7, 37.2 and 39.3 across the remaining bands out to the edge. One small patch faces the lens. Everything else faces green.
It is the finish that decides it, not the colour
How strong the reflection is comes down to the Fresnel term, the fraction of light a surface reflects rather than absorbs. Run the same sphere in the same set with only that number changed and the green it picks up goes 5.5 levels for a matte finish, 7.7 for satin or gloss paint, 24.6 for a semi-metallic surface and 34.1 for chrome. Dulling spray is not a cosmetic trick here; it is the difference between a tint and a wash.
Colour barely enters into it. A red helmet in the same set measures minus 30.7, because its own red still dominates what the camera sees, and the green never becomes the strongest channel. That also means the despill control cannot fire on it at all — it only acts where green is already above red and blue — but a red helmet is the one subject that does not need it.
Why it usually does not punch holes
The reflection has to swamp what the object looks like on its own before the keyer will delete it. That chrome sphere in the cyclorama reads a mean of (104, 178, 119): the green channel is right on the screen, but the red channel is at 104 because the studio lights are still lighting it, and the distance to the key works out at 0.468 against the 0.259 that would leave it untouched. It survives, and it survives the wrong colour.
Holes need a dark object with almost nothing of its own to lose. A dark glossy sphere in the cyclorama loses 0.8 per cent of its pixels; the chrome loses none; even a dark chrome ball forty centimetres off a flat screen loses none. What you get instead is a subject that is the right shape and roughly twice as green as any diffuse spill would have made it.
Flat screen or cyclorama is the whole variable
The same chrome ball in front of a flat screen one metre behind carries 8.5 levels of green, has 8.3 per cent of its pixels touched and loses 2.1 per cent outright. Inside the cyclorama it carries 34.1, because the screen stopped being one direction and became most of the hemisphere. Distance moves it the same way, since a flat screen shrinks in solid angle as you back off: 13.5 levels at half a metre, 8.5 at one, 3.2 at two and 0.1 at four.
So the question to ask on set is not how shiny the object is, it is how much of what it can see is green. A product on a flat panel two metres back is a completely different job from the same product sitting inside a cyc, and it is the second one that comes back looking like it was shot through a green filter.
The control people reach for is the one that destroys it
At the settings that actually keep the screen clean, nothing here is in danger: similarity 30 and 42 both leave the chrome ball untouched. At 63, 2.6 per cent of it starts to soften, and all of that is at 0.95 of the radius, out on the edge. That is a manageable amount of damage at the point where most people would stop.
Push it to 100 and it stops being edge trouble. Now 93.7 per cent of the sphere is touched, 39.8 per cent is deleted outright, and the mean radius of what got damaged is 0.70 — the hole has moved well inside the silhouette, and the only part still solid is the patch at 0.17 looking back at the lens. The screen is perfectly clean the whole time. You paid for it with the object.
Here despill is the right tool, which is unusual
Green sitting on a mirror is not the same thing as green bouncing onto skin. It is the screen, seen directly, and it should be replaced by whatever you put behind the subject — so clamping the green channel is closer to correct than it is anywhere else in this craft. Measured on the chrome sphere: 52.6 levels of green with suppression off, 34.1 at the default, 15.7 at seventy and minus 0.1 at a hundred.
Take it for what it is, though. The clamp does not know why the green is there, and on a neutral metal there is nothing of its own to lose, which is why full strength costs nothing here. On anything with a real colour of its own it will take that colour off too, and on a red helmet it never fires at all.
What to change before you press record
Dull the surface first. Dropping that Fresnel number from 0.90 to 0.05 — mattifying spray, a different finish, or just waiting for the polish to settle — takes the wash from 34.1 levels down to 7.7, which is the single biggest number on this page. If the object has to stay mirror-bright, the next lever is what it can see: opening the dark side of the set from a ten degree cone to a ninety degree one drops it from 35.4 to 20.9 and pushes what is left out to the rim, where it belongs.
Third, get it off the cyc. A flat panel two metres behind brings the same chrome ball down to 3.2 levels and 1.4 per cent of lost pixels, with the screen still keying perfectly. And if none of that is possible, leave the sliders alone: the default settings keep the whole object and hand it to you green, which despill can then take off, rather than deleting a third of it and handing you a hole.
Related questions
Does a shiny object really get holes keyed through it?
Rarely, and not at the default settings. A chrome sphere inside a cyclorama loses none of its pixels, because the studio lights still put its red channel at 104 and keep the distance to the key at 0.468. What it gets is colour: 34.1 levels of green on average. Holes appear when you open similarity — at 100, 39.8 per cent of it is deleted.
How much green does a mirror finish actually pick up?
Inside a wraparound set, 34.1 code levels on average above its own red and blue, peaking at 44.5, and 52.6 with spill suppression switched off. In front of a flat screen a metre behind the same object carries 8.5, and at four metres it carries 0.1.
Does spill suppression fix a mirrored subject?
Yes, better than it fixes anything else. On the chrome sphere it takes the green from 52.6 with the control off to 34.1 at the default, 15.7 at seventy and minus 0.1 at a hundred, because the green on a mirror genuinely is the background. It cannot help a red subject at all, since it only acts where green is above red and blue.
What is the cheapest fix on set?
Dull the surface: the measured difference between a chrome finish and a satin one is 34.1 levels of green against 7.7. If it has to stay glossy, flag the screen out of the directions it mirrors, or move it off the cyclorama onto a flat panel two metres back, which measures 3.2 levels and 1.4 per cent of lost pixels.