When is a blue screen the better choice than a green one?

The one number that decides it

One number decides it here: the distance in normalised RGB from a pixel to the key colour. At the defaults — Similarity 42, Smoothness 14 — anything within 0.161 of the key colour is fully transparent, anything beyond 0.259 is fully opaque, with a smooth fade between.

Switching from the green preset #00B140 to the blue preset #0047BB moves the centre of that sphere and nothing else: same radius, same fade, same two sliders. So "is blue better" reduces to "is blue farther from what I am shooting" — a thing you can measure rather than argue about.

Solid colours rarely fail; not one common subject colour we tested sits inside the window on its own. What fails is the half-covered edge pixel holding part subject and part screen, which sits at exactly half the subject's distance to the key colour. That puts the failure line for a pure subject colour at 0.518.

What blue actually wins at

Measurements below are on representative sRGB values. "Half edge" is the half-covered pixel, the one that actually gets eaten.

• olive jacket #6B6B3A — 0.502 from green, 0.672 from blue; half edge 0.251, inside the window on green
• dark skin #8D5524 — 0.669 / 0.812
• ash blonde #D8C078 — 0.877 / 1.006
• jet black hair #0F0F0F — 0.666 / 0.712
• charcoal suit #2B2B2B — 0.558 / 0.599
• denim jeans #3B5C7E — 0.473 / 0.343; half edge 0.237 and 0.171, fails on both
• navy blazer #1B2A4A — 0.541 / 0.470; half edge 0.235, fails on blue
• blue eyes #4A7FB5 — 0.577 / 0.365; half edge 0.182, fails on blue
• white shirt #F2F2F2 — 1.205 / 1.182, green safer

Tally: blue was the farther screen for 11 of the 16, green for 5. On the half-covered edge green lost two items and blue lost three — and denim lost on both, which is worth knowing before you buy a second roll of material.

Blue also has seniority. As one long-standing answer on Graphic Design Stack Exchange puts it, chroma keying "comes from way back in the days of film, when it was originally called Blue Screen". Green did not replace blue because it separates better — the table above says blue is usually the better separator. It replaced blue because it is brighter.

The brightness argument, measured

Brightness is blue's real cost, and it is quantifiable. The green preset sits at luma 0.515, the blue preset at 0.252 — log2(0.515 / 0.252) = 1.03 stops more exposure to bring blue to the same level under the same lamps. That is a stop of light you have to find, or a stop of noise you have to accept.

Brightness also shifts the key distance, slightly, and in both directions: for a neutral subject with no hue of its own, blue is farther below luma 0.527 (about 135 in 8-bit) and green is farther above it, with a margin peaking at 0.046 and falling to 0.023. Against hue margins of 0.170 for olive and 0.213 for blue eyes, that is a tie-breaker, not an argument. It only decides when the subject has no colour of its own.

What blue costs you: cleanup

This is the part that should actually decide it, because it comes from how this keyer is written rather than from photography folklore. In the code, the transparency mask is computed from the pixel's colour first; the spill suppression runs afterwards and writes only the green channel. Two consequences follow from that ordering.

• The spill slider can never change the shape of the cutout — the mask was already fixed before it ran.
• The rule it applies is green > max(red, blue). On a blue screen that is never true, so the slider does nothing at all to blue footage.

We modelled bounce as an additive fraction of the screen's own colour, normalising the two screens to equal reflected luminance — what you get when both are exposed to the same level. Reading is dE76 from the clean surface, before the slider, at the default 35, and at 100:

• light skin — green 10.28 / 10.28 / 10.28; blue 14.99 / 14.99 / 14.99
• dark skin — green 11.35 / 11.35 / 11.35; blue 17.55 / 17.55 / 17.55
• mid grey — green 11.24 / 7.60 / 3.01; blue 15.33 / 15.33 / 15.33
• jet black hair — green 14.07 / 9.38 / 3.15; blue 19.02 / 19.02 / 19.02
• white shirt — green 6.73 / 4.85 / 2.66; blue 5.74 / 5.74 / 5.74

On four of the five surfaces blue's leftover error is the larger one, and on all five it is the one you cannot reduce afterwards. The white shirt is the interesting case: blue starts cleaner at 5.74 against green's 6.73 and still ends worse, because green's error can be taken to 2.66 and blue's cannot be touched.

The slider is not free on green either. Because it drags green down to the level of the blue channel, a half-covered edge pixel on a white shirt goes from (121, 210, 153) to (121, 153, 153) at full strength — the edge turns teal. And on footage keyed against blue it can make things worse: ash blonde carrying heavy blue bounce measured 29.40 dE before and 30.49 after, because green was never the problem.

Deciding in sixty seconds

Shoot one frame, sample the largest three or four colour areas of your subject with the eyedropper, and compare each against both presets. Then:

• Anything large and green — olive, khaki, foliage — take blue. Olive is the biggest swing in the table at 0.170.
• A dark, neutral subject — black hair, charcoal, deep shadow — take blue, but only by 0.04 to 0.05. A tie-breaker, not a rescue.
• Denim, navy, or blue eyes — stay green. All three sit inside the window on blue once half-covered edges are counted.
• A subject close to the screen, shiny, or lit hard enough to throw visible bounce — stay green. Not because green bounces less, but because green bounce is the only kind you can remove here.
• Nothing applies — stay green. It costs one stop less of light.

One caveat: the colours above are representative values, not samples from a photograph, and the bounce levels are a modelled scan. What transfers to your shoot is the method — sample your subject, halve the distance for the edge, compare with 0.259.

Related questions

Open the green screen remover