What do you do when your key effect has no screen gain control?

In most cases, nothing — because what people want from screen gain is usually just a less aggressive edge, and that is exactly what the key window controls already give you. Screen gain does something much more sweeping: it multiplies every alpha in the result, including the pixels inside your subject that should be a solid 1. At gain 0.80 the subject itself drifts up to 15.94 dE away from its own colour once composited, because it has started letting the new background through. We measured every substitute against it and none can reproduce it, because they work on a different axis: the window controls decide which pixels are semi-transparent, while gain decides how transparent the semi-transparent ones actually are.

What the control actually does

Before deciding whether you miss it, be clear about what it does. The control people mean by screen gain scales the matte: it takes every alpha value the keyer produced and multiplies it. alpha_out = alpha_in x g, with values clamped back into 0 to 1.

That is a different operation from anything the other controls do, and the difference is easy to miss because on a test frame the results look similar. Turn it down and the green recedes. So does lowering the key tolerance. But they got there by different routes, and one of them charges you for it.

The rest of this article is arithmetic on that definition. If your software implements the control differently, the numbers below will not transfer — but the distinction between the two axes will.

The part nobody warns you about: your subject goes translucent

Here is the cost. A pixel in the middle of your subject has coverage 1 and its alpha should be exactly 1. Scale it by 0.80 and it becomes 0.80, and 0.80 means twenty percent of whatever you composite behind it is now showing through your subject.

That is a colour error across the whole body, not just at the edge. Measured as CIE76 dE from the colour the subject was supposed to be, at gain 0.80:

denim blue over warm orange — 15.94
blonde hair over deep blue — 15.45
white shirt over deep blue — 14.50
dark skin over white — 14.35
light skin over deep blue — 13.93
white shirt over warm orange — 12.48
denim blue over white — 4.05
white shirt over white — 1.05

Two things worth noticing. First, the size of the error depends mostly on how different your new background is from your subject, not on the key itself — a white shirt over a white background loses 1.05 dE because there is barely anything to bleed through, while the same shirt over deep blue loses 14.50. Second, the worst cases land at 15 to 16 dE, which is not a subtle shift.

At gain 0.95 the same figures are roughly a quarter of that, between 0.26 and 3.82. So the damage scales fast, and it starts the moment you move off 1.00.

And here is what it does to the edge

Scaling every alpha also means the edge loses mass. Scanning a synthetic edge from zero coverage to full coverage, with the default key window, here is what gain does to the count of genuinely semi-transparent pixels and to the average alpha across the edge:

white shirt — 8 semi-transparent pixels at gain 1.00, 87 at gain 0.90, average alpha 0.8208 falling to 0.7387
light skin — 9 at gain 1.00, 84 at 0.90, average 0.7871 to 0.7084
blonde hair — 11 at 1.00, 82 at 0.90, average 0.7580 to 0.6822
dark skin — 15 at 1.00, 75 at 0.90, average 0.6701 to 0.6031
denim blue — 20 at 1.00, 66 at 0.90, average 0.5555 to 0.4999

The jump from 8 to 87 on a white shirt is not the edge getting softer. Most of those new semi-transparent pixels are inside the subject, where alpha used to be a clean 1. That is the cost from the previous section showing up in a different measurement.

So gain buys you a receding green edge and charges you an increasingly translucent subject. Whether that is a good trade depends on the shot, but it is never free, and it is not what most people think they are reaching for.

Why no other control can reproduce it

We tested every substitute we could think of against a gain of 0.80 on a denim edge, measuring the average absolute difference across the whole alpha curve, and watching whether the subject itself got weakened.

similarity 42 (default) — difference 0.1111, subject untouched
similarity 38 — difference 0.1530, subject untouched
similarity 34 — difference 0.1949, subject untouched
similarity 30 — difference 0.2368, subject untouched
smoothness 14 (default) — difference 0.1111, subject untouched
smoothness 30 — difference 0.1158, subject untouched
smoothness 50 — difference 0.1400, subject untouched
choker 0.05 — difference 0.1126, subject untouched
choker 0.10 — difference 0.1156, subject untouched
choker 0.20 — difference 0.1236, subject untouched
gain 0.80 itself — difference 0, subject weakened to 0.800

Every single substitute leaves the subject at a full alpha of 1.000. Only gain takes it down. That is the whole finding, and it is a categorical one rather than a matter of degree.

The reason is that they operate on different axes. The key window controls decide which pixels are semi-transparent and how wide that band is. Gain decides how transparent the semi-transparent ones are. No amount of moving a threshold can change a value, which is why turning similarity down moves you further from gain rather than closer to it — 0.1111 becomes 0.2368, because a narrower window makes the edge steeper while gain makes it softer. They are pulling in opposite directions.

What to reach for instead

Almost always, what you actually wanted was a gentler edge, and two controls cover that without touching your subject.

For how much gets removed, use the key window. Lowering similarity narrows it, which keeps more of the semi-transparent band. The trade is honest and visible: more of the screen survives as partial transparency, so you may see haze where you used to see a clean cut.

For how soft the boundary is, use smoothness. It sets the width of the semi-transparent band — 0.098 at the default here — without moving where the cut falls. That is the control to reach for when the edge is the right shape but too hard or too soft.

For a fringe that is the wrong colour, use spill suppression. That is a colour operation, not an alpha one, and no amount of alpha scaling will fix green that is still in your RGB.

If you genuinely need the whole subject to sit at less than full opacity, that is a composite decision — do it on the layer opacity in your editor, where you can see it, rather than buried in the keyer where it looks like an edge fix.

The case where no control saves you

There is one situation where the missing control is not the problem at all, and it is worth naming because it is the most common reason people go looking for it.

If your screen is unevenly lit — a hot spot behind one shoulder, a falloff toward the floor — then the key colour is not the same colour across the frame. A keyer has one key colour. No single setting fits everywhere, so every setting is a compromise, and it looks like you need a stronger control.

You do not. Screen gain is global. It cannot be high on one shoulder and low on the other. Whatever it does, it does to the whole frame, which is exactly the wrong shape for a local problem.

That one is fixed at the lighting stage, not the keying stage: separate the subject from the screen, light them independently, and measure the falloff before you shoot. Our piece on uneven lighting covers the numbers.

Why this tool deliberately leaves it out

GreenBG Remover exposes the key window — similarity and smoothness — and spill suppression. It does not expose a global alpha gain, and that is a choice rather than an oversight.

The reasoning is the third section of this article. A control that scales every alpha in the result will, by construction, take pixels that were correctly fully opaque and make them partially see-through. Every measurement we ran says the same thing, and the failure mode is quiet: it does not look like an error, it looks like a slightly soft image.

Giving you two controls that only move the boundary is more restrictive, and it means that one particular look is harder to reach. That is the trade we made.