Colour Inconstancy (CII)

One colour, one surface, and it changes as the light changes. Some colours barely move; others are unrecognisable between a light booth and a supermarket aisle. Which kind you have chosen is a property of the colour, and you can know it before anything is printed.

The measurement, in one line

Compute the colour under a reference light. Compute it again under another. The distance between the two answers is its Colour Inconstancy Index — and because a real observer adapts to the light they are standing in, a proper CII adapts too, rather than reading off the second set of numbers.

What the browser shows. Changing the illuminant here recomputes each sample under the new light without a chromatic adaptation step — the same arithmetic the library has always used. It is the right number for “what does the instrument say under that lamp”, and it is close enough to rank colours by how badly they move. A CII with full adaptation is a figure the desktop application reports.

How far colours actually move

Three colours under four lights Rusty Red moves 9.6 dE00 between D50 and tungsten; Soft Conch, a typical colour, moves 2.8; the neutral Ashen Nickel moves 0.3. All computed from the library spectra, all inside the sRGB gamut under every light. D50daylight, the referenceD65cooler daylightF11narrow-band shopAtungsten Rusty Red 7R3.50 moves the most reference ΔE00 2.7 ΔE00 5.8 ΔE00 9.6 Soft Conch 2R4.68 a typical colour reference ΔE00 1.2 ΔE00 1.1 ΔE00 2.8 Ashen Nickel N.50 barely moves reference ΔE00 0.1 ΔE00 0.2 ΔE00 0.3 Three colours under four lights The same comparison, narrower. D50D65F11A Rusty Red moves the most ref 2.7 5.8 9.6 Soft Conch a typical colour ref 1.2 1.1 2.8 Ashen Nickel barely moves ref 0.1 0.2 0.3

Three samples from the library, computed from their own spectra under four lights and rendered here in sRGB — all three stay inside the screen’s gamut under every light, so what you see is the movement and not a clipping artefact. The numbers are ΔE00 from the D50 column. One colour is unrecognisable by tungsten, one shifts a little, one does not move at all; nothing was done to them except changing the lamp.

Every sample in the library, computed under each light, against D50 at 2°:

Change the light to…Median 95%WorstWorst sample
D65 — daylight, cooler1.443.788.607V4.28
F11 — narrow-band shop1.605.9315.806B2.28
A — tungsten3.388.1717.607V5.23
F2 — cool white fluorescent3.817.9410.527B4.55

ΔE00, M2, 2° observer, all 4253 samples.

Two things are worth taking from that table. The median is not small: a typical colour moves 1.4 to 3.8 ΔE00 when the lamp changes, and the library’s own neighbouring swatches are 2.5 apart. And the worst is far worse than the median — the distribution has a long tail, and everything expensive lives in it.

The tail is not random

The colours that move most are saturated blues, violets and deep cyans — 7V4.28, 6B2.28, 7V5.23, 7B4.55 turn up at the top of every column. Their spectra have most of their energy where lamps differ from each other most: a fluorescent tube has spikes there, tungsten has almost nothing there, and daylight is smooth.

Yellows and warm reds sit near the bottom of the same lists. Not because they are simpler, but because the red end of the spectrum is where lamps happen to agree.

The practical consequence. A brand colour in the saturated blue-violet region will need its lighting specified. One in the warm end mostly will not. That is not a rule of thumb — it is what these four columns say.

Using it while you choose

Pick the light before you pick the colour. If the product lives on a shop shelf, D50 in a booth is not the condition that matters. Set the Illuminant control to something like it and look at the candidates there.

Compare neighbours. Two colours a step apart in the book can behave very differently as the light changes. Where a design has any freedom, the steadier neighbour is usually free.

Check both ends, not the average. A colour that is fine under daylight and fine under tungsten is fine. A colour with a good average and one terrible lamp is a reprint waiting to happen.

Not the same as metamerism

Inconstancy is one colour changing. Metamerism is two colours that agreed under one light disagreeing under another. A colour can be perfectly constant and still be half of a bad metameric pair, and a very inconstant colour can still match its partner everywhere — if the two share a spectrum.

Both are computed from the same thing, and only from that thing: the spectrum.

For the thorough. The full figures behind this page are on Inconstancy: the full matrix.

Read next