showstack

The distinction

Adding light, and taking it away

Additive mixing is what emitters do. Two lamps pointed at the same wall each deliver their own spectrum and the wall receives the sum. You start at black — no light — and every source you bring up adds. Red plus green is yellow, which is nonsense with paint and obvious with light: the surface is receiving long wavelengths and medium wavelengths at once, and the eye has no way to tell that apart from a genuine yellow.

Subtractive mixing is what filters do. One white source passes through things that each remove part of the spectrum. Removal is multiplication, not addition: a filter passing 40% of the red and a second passing 50% of it gives 20%, not 90%. You start at white and take away, and if you stack enough you get black.

The practical consequence follows immediately and it catches everybody once. On an additive fixture, saturation is free — a deep blue costs you the red and green emitters, which were not doing much for a blue anyway. On a subtractive fixture, saturation costs output. Every step deeper is another chunk of the spectrum thrown away as heat, which is why a CMY unit at a deep congo is a fraction of its own datasheet and why the same colour on an LED unit next to it looks like a different fixture.

The question everyone asks

A shadow is not an absence of light

Point a warm source at an object from one side and a blue one from the other, and you get two shadows, one blue and one orange, and neither of them is grey. People find this uncanny, and the explanation is almost disappointingly mechanical.

A shadow is not darkness. It is the set of sources that still reach the surface. The object blocks the warm source in one direction, so that patch of wall is lit by the blue source and by nothing else — so it is blue. Exactly, precisely blue: the same colour the blue source is, at whatever level it arrives. The other shadow is orange for the same reason. Where neither is blocked you get both, which is why the wall between the two shadows is brighter than either shadow and closer to white than either source.

That is the whole physical story, and you can watch it below. There is a second reason shadows read coloured, and it is perceptual rather than physical: the visual system continuously re-normalises to whatever illuminant dominates, so under a strong warm key a genuinely neutral shadow is seen as blue because it is bluer than the thing your eye has adapted to. Both effects are real and they compound, which is why a warm key with an ordinary white fill produces shadows people insist are blue on film that measures neutral.

Inside the fixture

Flags in a beam, or a handful of emitters

Two architectures do the same job and they fail differently, which is why a rig with both in it never quite matches.

A CMY unit is subtractive. One white source — discharge or a white LED engine — and three dichroic flags that slide progressively into the beam, each removing one third of the spectrum. Dichroic means the filter reflects what it does not pass rather than absorbing it, which is how it survives being in a beam that would burn a gel to ash in seconds. The mixing is continuous and smooth, the pastel end is beautiful, and the deep end is dim.

A multi-emitter LED unit is additive. Several colours of LED, driven independently, summing in a mixing chamber. Three emitters can technically reach a lot of colours, and the reason nobody ships three is that the spectrum between them is empty: an RGB source has almost no energy in the yellows and cyans, so anything reflecting there goes dead. So manufacturers keep adding emitters — white for output and pastels, amber for warmth and skin, lime for the enormous luminance gain green-yellow gives you, cyan and royal for the gap at the blue end.

The consequence for a programmer is that the same numbers do not mean the same thing on two fixtures. A colour picked on a CMY unit and sent to an LED unit lands somewhere else, because the fixtures got there by different routes through different spectra, and any calibration is a fit rather than an equivalence.

The costume

An object can only reflect what it is given

A red costume under a saturated green wash does not go dark red. It goes black — and the first time it happens in a tech run somebody blames the fixture.

What you see from a surface is its reflectance multiplied by the source’s spectrum, wavelength by wavelength. A red dye reflects long wavelengths and absorbs the rest. A saturated green LED emits almost nothing in the long wavelengths. Multiply the two and there is nothing left to come back. The costume is not dark; it is receiving no light it is capable of returning.

This is why reflectance and transmission are the same arithmetic and the mixing tool answers both with one calculation: put a source and a costume in subtractive mode and you get exactly what comes off the fabric. And it is why the fix is never “more light”. Four times the green wash returns four times nothing. What the costume needs is a source with energy where it reflects, which is a design decision made at fixture-selection time, not a level change made at 11pm.

The disagreement

Metamerism, and why the camera says otherwise

Two lights can look identical and be made of completely different spectra. The eye has three cone types, so it collapses an entire spectrum into three numbers, and a great many different spectra collapse to the same three. Two such spectra are metamers: indistinguishable to a human, and not remotely the same physical thing.

That is a convenience right up until something else looks at them. A camera has its own three sensitivity curves and they are not the eye’s, so it collapses the same spectra differently — and two sources that matched by eye come out visibly apart on the feed. This is the mechanism behind the classic argument in a tech: an LED wall and a lighting fixture matched carefully by eye, clashing on the broadcast output, with everybody certain the other department has it wrong. Both are seeing what they say they are seeing.

It also runs the other way, across objects rather than sources. Two pieces of costume that match under the workroom’s fluorescents can separate under a narrowband LED wash, because they were only ever metamers under one spectrum. Matching fabric under the light it will actually be seen in is not fussiness, it is the only test that means anything.

Asking for better

CRI, R9 and what TM-30 adds

CRI is one number summarising how faithfully a source renders a set of test colours against a reference. It is old, the sample set is pale, and averaging eight pastel patches hides exactly the failure that matters on a stage: R9, the deep red sample, is not in the average at all. A source can post CRI 95 and still be poor at R9, which shows up immediately as bad skin, dead reds, and a flat look nobody can fix with gel.

TM-30 replaces the summary with two numbers and a picture. Rf is fidelity — how close to the reference, on 99 samples rather than eight. Rg is gamut — whether the source expands or compresses saturation relative to the reference. High Rf with low Rg is a source that is accurate and flat. High Rg is a source that makes everything look more colourful than it is, which some people like and which is still a distortion.

What to actually ask a manufacturer for: Rf and Rg, R9 stated separately from CRI, and the spectral power distribution if they will publish it. And then look at the fixture next to a face, because the reason all of these metrics keep getting replaced is that none of them fully predicts what skin does.

The other half

Colour psychology, and which parts of it survive

Colour is the fastest emotional lever on a stage, and it is also the subject of more confident nonsense than anything else in this trade. So it is worth separating what is physiology from what is culture from what is marketing.

What is genuinely physiological. The eye has a fifth photoreceptor beyond the three cones and the rods — intrinsically photosensitive retinal ganglion cells, most sensitive around 480 nm, in the blue. They do not contribute much to what you see; they feed the circadian system and measurably affect alertness. Short-wavelength-heavy light in the evening delays sleep and raises alertness, and warm low-level light does the opposite. That is a real, replicated effect with a known mechanism, and it is why a get-out under 5600 K feels different from one under 2700 K in a way that is not imagination.

Two more that are solidly physiological. Saturated deep red flickers differently and is treated separately in the photosensitivity guidance for that reason, which the access page covers. And the eye cannot focus red and blue at the same distance — the lens is dispersive, so a saturated red and a saturated blue at the same depth resolve at slightly different focal planes. That is chromostereopsis, it is why deep blue text on a red ground appears to float, and it is a genuine optical effect rather than a perceptual quirk.

What is learned rather than innate. Nearly all specific hue-to-meaning mappings. White is mourning in parts of Asia and celebration in parts of Europe. Red is danger on a control panel, luck at a Chinese wedding, and left-wing in one political vocabulary and right-wing in another. These associations are strong, they are real, and they are local — a lighting choice that reads as ceremonial in one venue reads as funereal in another, and no amount of theory tells you which without knowing the room.

What does not survive contact with the evidence. The popular hue-to-emotion tables. “Blue is calming, yellow is happy, green is balanced” is a marketing artefact rather than a finding: the effects are small, inconsistent between studies, heavily dependent on saturation and brightness rather than hue, and confounded by exactly the cultural associations above. This does not mean colour has no emotional effect — it plainly does — it means the effect is not carried by hue alone in the way the tables claim.

Where this goes next

The calculators on this page

Colour mixing and shadows does both operations in linear light and shows what each source’s shadow will be. Mixing colour temperatures works the mired arithmetic and warns about the off-locus green shift. Gel correction picks the filter that gets one source to another. And how a colour becomes a number is the other half of this — what happens once the light has been turned into data.