Why do batch product shots still need hand fixes after a green screen reshoot?
Because a batch has to run one setting across every frame, and the two things that setting has to satisfy are in conflict. Each colourway has a ceiling above which the keyer starts eating its outline — measured on a 70 IRE screen: teal 50, forest green 58, navy 62, black 66, red 120, tan 123, yellow 146, white 168. A contact shadow on the screen at half its lit level needs similarity 61 to clear, which is above the ceiling for teal and forest green; at a third of screen level it needs 83, which also breaks navy and black. So the reshoot removes the backdrop and hands back every contact shadow plus the two or four colourways that no single setting can serve.
What the reshoot actually buys you
A green screen reshoot solves one problem completely: the thing behind the product is now a known, flat colour, so it can be removed by arithmetic instead of by hand. That is genuinely worth doing, and it is the largest single saving available.
What it does not solve is that every frame still contains two independent decisions — how much of the frame counts as background, and what to do about the shadow the product casts on it. In a single hero shot you make those decisions once, by eye, and move on. In a batch of two hundred you make them once and live with the answer on every frame, including the ones where it is wrong.
Every colourway has a ceiling
A colour key deletes a pixel when it is close enough to the key colour, and each product colour sits at its own distance from it. Take eight colourways of the same product on the same screen, keyed with the colour picked from the screen itself, and raise the similarity until the first rim pixel is deleted — a rim pixel being one that is more than half product, which should never disappear:
white 168, yellow 146, tan 123, red 120, black 66, navy 62, forest green 58, teal 50. Those are ceilings, not preferences. Above them the outline starts to go: for the forest green SKU, 14.4 per cent of its rim pixels are deleted at similarity 60, 31.0 per cent at 70, and 49.4 per cent at 80. That is the product visibly losing pieces of its own edge.
Individually none of this is a problem, because individually you would pick a setting under each ceiling. The batch is where it bites.
The contact shadow is what breaks it
A product sitting on a screen blocks some of the light reaching it, and the patch under and around the product is the same green at a lower exposure. That is not the same colour as the rest of the screen, and the keyer treats it accordingly.
The failure is not gradual, it is a cliff. Model the shadow as the screen at a fraction of its lit level: at 70 per cent the shadow pixel sits 0.142 from the key and is removed completely, alpha 0.000. At 50 per cent it sits 0.257 away and comes back at alpha 0.999 — fully opaque. Composited on white, which is where catalogue images end up, that reads (0,168,60): a solid mid-green shape where the shadow should be, 332 code levels away from the white it is supposed to sit on.
It is also not fixable with the despill controls, which is worth knowing before you plan a cleanup pass. On a shadow pixel green is the largest channel, so the sliders do fire, and they take (0,167,60) to (0,123,60) — still 347 levels of smudge on white. Despill can change the shadow's hue; it cannot make it transparent.
Where the two requirements cross
Now put the two constraints on the same axis. Clearing a shadow needs a similarity of about 200 times its distance from the key plus 10. Working down from a light shadow to a deep one, and comparing each requirement against the eight ceilings:
A shadow at 90 per cent of the screen's level needs 19, at 80 per cent needs 28, at 70 per cent needs 38, at 60 per cent needs 49 — all of which sit below every ceiling in the set, so all of them clear cleanly. At 50 per cent the requirement is 61, which is above forest green at 58 and teal at 50: those two break. At 35 per cent it is 83, which additionally breaks navy at 62 and black at 66. At 20 per cent it needs 110, and the control stops at 100.
That is the whole answer in one table. A soft shadow costs nothing. A normal contact shadow costs you two colourways. A deep one costs you four, and a very deep one cannot be keyed out at all.
Two smaller things that make it worse
The first is the key colour itself. Pick it from the screen and a clean screen needs similarity 21. Leave it on the #00B140 preset while lighting the screen to 70 IRE — the combination from the previous section — and the screen needs 64, which is above navy, forest green and teal on its own. Three of eight colourways break before the shadow is even in the picture, purely because the preset is a hue at one particular exposure.
The second is consistency, which is a requirement in a catalogue and not in a single shot. Even if you tuned every frame optimally, a set in which one colourway has a crisp edge and another has a soft one reads as sloppy on a grid page. So the honest target is the best single setting for the whole batch, which means accepting a setting that is worse than optimal for most frames in order to keep them uniform — and then checking by hand that uniform actually looks like uniform.
What to do instead
The lever with the most effect is the shadow, and it is a lighting problem. Get the shadow no darker than about 60 per cent of the screen's lit level and it clears at a similarity every colourway survives. In practice that means lifting the product off the screen on a stand, or adding a dedicated light to fill the shadow region, so the shadow stays a shading of the screen rather than a different colour. Then one setting does the whole batch.
If you cannot relight it, decide deliberately which way to fail. Either raise similarity and rebuild the outlines of the two or four colourways that break, or keep similarity low and rebuild the shadow on every frame. The first is fewer frames; the second is more predictable work. Both are hand work, and knowing which one you have chosen before you start is worth more than discovering it at frame 180.
And if the product range contains a green or teal colourway, test that one first. It is the one that sets the ceiling for everything else.
Related questions
Why can't I just use one key setting for the whole batch?
You can, and that is the right thing to do for consistency — but the setting has to clear the contact shadow as well as the screen. A shadow at half the screen's lit level needs similarity 61, while teal can only survive up to 50 and forest green up to 58, so those two lose part of their outline at the setting that clears the shadow.
Can the despill sliders remove the green shadow?
No. They fire on a shadow pixel, because green is its largest channel, and they take (0,167,60) to (0,123,60) — still 347 code levels of smudge when composited on white. Despill changes the shadow's hue, not its opacity. The shadow has to be lit away or removed by hand.
How dark can the contact shadow be before it becomes a problem?
About 60 per cent of the screen's lit level. At that depth it needs similarity 49 to clear, which is below every colourway ceiling in the eight-SKU test. At 50 per cent it needs 61 and two colourways break; at 35 per cent it needs 83 and four break; at 20 per cent it needs 110, past the end of the control.
Does it matter which colourway I test first?
Yes — test the green or teal one. Those sit closest to the key colour and set the lowest ceilings: teal 50 and forest green 58, against white 168 and yellow 146. Whatever setting works for the green colourway works for the rest; the reverse is not true.