When is a plain backdrop better than a green screen you can't light properly?

When the room cannot put the cloth far enough behind you, and when you do not actually need the backdrop removed. On a 960x540 webcam head-and-shoulders frame the cloth worked from 0.30 m of clearance: similarity 55 removed every cloth pixel and cost no part of the subject. At 0.25 m your own shadow became a full umbra, the darkest cloth pixel went black at 0.78 from the key, and clearing every last pixel needed similarity 160, which took 98.42% of the subject with it. A plain wall has no radius to fall inside and gives back colourless light, so in a room that tight it is the better backdrop — provided you can leave it in the shot.

How much backdrop a webcam frame actually needs

A laptop camera 0.7 m in front of you frames about 1.00 x 0.56 m at your shoulders. Anything behind you has to fill that frame at its own distance, so the backdrop grows with the clearance you give it: 1.36 x 0.76 m at 0.25 m behind you, 1.71 x 0.96 m at 0.50 m, 2.43 x 1.37 m at 1.00 m, 3.14 x 1.77 m at 1.50 m and 5.29 x 2.97 m at 3.00 m. Add the 0.7 m in front of you and 0.4 m behind the laptop and the room has to be 1.35, 1.60, 2.10, 2.60 or 4.10 m deep.

That is the part people find out too late. The 1.5 m collapsible panel most people own can only sit about 0.30 m behind you at this framing before its edges walk into shot, which is exactly the distance where things start to break. Lighting it costs more as it grows as well: with two fixtures pushed as far back as the room allows, holding the same cloth exposure took 1.00x, 1.10x, 1.61x, 2.38x, 3.37x and 6.01x the output across those same distances.

A colour key has to satisfy two distances at once

This keyer measures each pixel's distance to the key colour in normalised RGB and turns a pixel fully transparent once that distance is at or below similarity/200 minus smoothness times 0.0035; at the default 42 and 14 that radius is 0.161. Two numbers decide whether any single setting can work. The worst cloth pixel has to fall inside the radius, which needs a similarity of at least 200 times its distance plus 9.8. The nearest subject pixel has to stay outside the far edge, which caps similarity at 200 times its distance minus 0.049.

Put together, one global setting exists only if the gap between those two distances is at least twice the soft band, 0.098 at the default smoothness. On the green cloth the nearest subject pixel sat at 0.4974, 0.4923, 0.4872, 0.4899, 0.4872 and 0.4867 across the six distances — barely moved. Everything that changes is on the cloth side, and that is where the room shows up.

The line is a shadow, and it falls at about 26 cm

What fails at close range is not the evenness. The lit part of the cloth was fine everywhere: its worst pixel needed only similarity 47 at 0.25 m. The failure is the shadow, and specifically whether it is a penumbra or a real umbra where both lamps are blocked at once. With lamps 0.80 m either side of the axis and 0.70 m in front of you, a torso of half-width 0.215 m casts an umbra that reaches w times a over L minus w, which is 0.215 x 0.70 / 0.585 = 0.257 m behind you.

The measurement lands on that number. At 0.25 m the darkest cloth pixel was effectively black, 0.7766 from the key, and shadow of some kind covered 32.2% of the cloth. At 0.30 m the darkest was 0.2585 and the coverage was 30.7% — the area barely changed, the depth did. Because every cloth pixel has to fall inside the radius, that single darkest pixel sets the price: similarity 160 at 0.25 m, against 55 at 0.30 m, and 160 takes 98.42% of the subject with it while 55 costs nothing.

What the cloth does to you while it is up

A backdrop is also a light source, and a green one only gives back green. On this frame the cloth supplied 20.9% of the subject's direct illumination on average at 0.25 m and 64.4% at the worst pixel; even at 3.00 m it was still 12.0% and 48.0%, because the cloth always fills the frame and always stays in view. Measured against the same subject with no backdrop at all, green excess ran +8.1 code levels on average, +19.0 at the 95th percentile and +21.0 at the worst at 0.25 m, falling to +4.8, +10.0 and +13.0 at 3.00 m.

Neither spill control touches it. The despill limiter only fires where green is already the strongest channel, and on skin, hair and shirt red or blue is ahead of green, so on the subject's own pixels the RGB distance from the no-backdrop reference was 9.3 before suppression and 9.3 after running both sliders to 100. Moving the cloth further back is the only lever that reduces that bounce, and it is the one a small room does not have.

What a plain wall gives back instead

Against the same subject and the same two fixtures, a mid-grey wall supplied 12.1% of direct illumination at 0.25 m — less, but not the point. It supplies it equally in all three channels, so green excess moved by -1.8 code levels on average, 0.0 at the 95th percentile and +1.0 at the very worst, against the cloth's +21.0. Total RGB displacement was 10.0 for the wall and 9.3 for the cloth: the wall is not giving back less light, it is giving back colourless light.

It also needs nothing built for it. The wall is lit by the room you are already sitting in; the cloth has to be bought, hung, and lit by its own fixtures whose output climbs from 1.00x to 6.01x as you push it further back. And nothing about the wall has to fall inside a radius, so its evenness, its corners and the shadow you cast on it are all somebody else's problem.

You cannot just key the wall instead

If you want the backdrop gone, the wall is not a substitute — it is worse, at every distance. The nearest subject pixel sat 0.2046, 0.1965, 0.1939, 0.1930, 0.1930 and 0.1914 from the wall's own colour, while the worst wall pixel sat 0.98, 0.32, 0.31, 0.32, 0.33 and 0.20 away. The gap stayed negative throughout, from -0.78 to -0.01, against the 0.098 a setting needs. Skin and mid-grey are simply too close together in normalised RGB.

Pushing the sliders confirms it. Making the wall completely transparent took similarity 200 at 0.25 m, which deleted 100% of the subject, 65 at 1.00 m with 74.86% of the subject gone, and 40 at 3.00 m still costing 3.04%. So the honest choice is binary: either keep the wall in the frame, or put something behind you whose colour is genuinely separable. A wall you half remove is the worst of both.

The decision, in one measurement

Measure the clearance between your shoulders and the backdrop. If it is 0.30 m or more and you can get light onto the cloth, use the cloth: similarity 50 to 55 cleared it completely in this frame with zero subject loss at every distance from 0.30 m to 2.00 m. Below roughly 0.26 m you are inside your own umbra, and no setting rescues it — you move the desk, put the camera on a longer lens from further back, or you stop trying to remove the backdrop.

If the backdrop does not need to come out at all, which is the usual case for a call, a lesson or a recording, the wall wins on every axis that matters here. It costs at most 1 code level of green against the cloth's 21, it needs no fixtures and no support, and there is no radius it has to satisfy. The cloth only beats it when you actually want the transparency, and then it needs room.