Color Wheels Are Wrong

The article was clearly written by an enthusiast, but is a bit muddled. I see nothing in what he has written that supports the idea that Color Wheels are wrong.

Color wheels are, in essence, highly abstracted slices across the equator of a color space. Only two color wheels are commonly used: the RGB and the RYB (aka the artists color wheel). Color from pigments and color from light are two wildly different things, but easy to confuse. For its absence of green as a primary, CMYK can be understood as a version of RYB. Black is used not just because it is cheap but also to add opacity to the inks (printing inks are more like dyes... very thin and transparent).

The color opposites the author describes are red/cyan in RGB and red/green in RYB.

One property of these opposites is that they mix to neutral. For this, the RBG pair works best. The other property is that they are perceptually antagonistic, for which the RYB pair works best.

Interestingly, the perceptually antagonistic pair set (red/green, yellow/purple, blue/orange) were spoken about well before Newton placed these colors in a circle. Leonardo Da Vinco refers to them as 'retto contrario', literally exactly opposite.

The author asks 'why a wheel'? which is a good question, and the answer he gives is reasonable. But... Newton was the first to employ this device, and he did so almost certainly because the harmony structure of the musical scale was commonly visualized as a wheel, and Newton was keen to draw an analogy between music and color and harmony and the music of the spheres and god etc etc. He may have been the first physicist, but he was also an alchemist.

Color is a frustrating beast... so many different ways to understand the same thing: perception, physics, aesthetics, chemistry etc. Truly the seven blind men and the elephant.

> Ask any artist to explain how color works, and they’ll launch into a treatise about how the Three Primary Colors—red, blue, and yellow—form a color wheel

I doubt that. More likely that they look at you akwardly or say a joke and try to carry on.

Some artist couldn’t care less about colours. A sculptor thinks much more about shapes, a 3d animator thinks much more about motion.

A writer or poet cares a lot more about what connotations different colour descriptors carry. “The sky above the port was the colour of television, tuned to a dead channel.” doesn’t quite hit the same as just calling it grey, for example.

Painters have a much more intimate connection with colour. But instead of thinking in primary colours they will be much more familiar with the mixing of the pigments they use. (Cadmium red, cobalt blue, burnt sienna, etc) A print artist or someone designing figurines or design objects would think in terms of pantone colours. Someone doing thread painting with embroidery floss will think about DMC colour codes.

And then of course there are colourist working on movie and TV productions who would know all the article describes and more. It is their job to know.

I don’t know why an otherwise quite okay article has to start with this image of the dum-dum arist “launching into a treatise” of kindergarden level colour understanding. It feels a bit degrading, as if the author has low opinion of artist. (And certainly not considering the full palette of the arts.)

> Black is included as a money-saver

Since when were printer manufacturers concerned about saving the end user money? lol.

Black is included because if you mix typical dye based inks you do not get a black, you get this greenish looking thing that is clearly not quite black. A few here are old enough to remember when early color inkjets could only hold either the tricolor cartridge or the black cartridge (ex HP 500C).

> Some of those frequencies we detect with our eyes, and the frequency determines its color

This is common misconception. It's not singular frequency of light that determines color, but the entire distribution of intensity over visible light spectrum.

And it is not one to one mapping from spectra to color either. In theory there are almost infinite spectra for each color.

So mixing red and green lasers might produce something appearing yellow(ish) despite the spectrum not containing any light corresponding to spectral pure yellow wavelengths.

As for the "art primaries", afaik it's not so wrong of a color model; it just is a reflection of common paints/pigments operate. Painters can not get nice spectrally pure pigments for primaries, not now and even less so historically, and so the color space of paints is significantly smaller than the entire human vision color space.

This topic has been written about in more depth by people who didn't try to work it out by themselves.

The classical artists color wheel is based on pigments. Printers use dyes. Screens use light. That's the whole reason why the primaries are different. The wheels are just tools.

The way traditional artists think and thought about colors is definitely not in the shape of color wheels.

Traditional artists think in terms of palettes and of mixtures of colors on the palette. They also think about colors in relation to each other. They think of layering, and about the perception of the eye.

The point is, when you discuss colors as a multi-dimensional coordinate space, you already lost (from the perspective of the painter).

Painters will explore the medium they are working in for its capabilities. Is "dark on bright" (watercolors) or "bright on dark" (soft pastel) the way to go? Can I dilute and perform washes (watercolor), or will the medium break down (acrylic).

I can only recommend James Guerney's "Color and Light" book to get impressions.

Then again, if you do digital art to scren, or if you are printing things from a computer, you need to anticipate how the devices render your colors. This is where color spaces and coordinate systems for colors and how to translate between them becomes relevant.

> Every seven-year-old kid in America is taught that “the opposite of red is green” and “the opposite of blue is yellow.”

In Europe as a kid I was taught in art class the following opposite pairs instead:

red-green

blue-orange

yellow-purple

And these pairs were each said to make brown when mixed (I know, it's very different compared to RGB or CMY where the neutral color is gray)

Color is hard, but this article doesn't do it justice. Wheels aren't broken, perception is just weird and filled with all sorts of fun non-linearities.

The article begins to touch on color spaces, which is where I'd suggest curious folks to begin if they're interested in color.[0] The article fails to note the fact that MANY other color spaces have been invented in attempts to address the problems it mentions![1] OKLab is a modern one that may be familiar to folks here. Scroll down to some comparisons vs other spaces, it's a nice uniform circle again![2] (OK, maybe not entirely uniform. Color is hard.)

Tangent: Why the heck are there alpha-blended images in an article about color!?

[0] https://en.wikipedia.org/wiki/CIE_1931_color_space

[1] https://en.wikipedia.org/wiki/List_of_color_spaces_and_their...

[2] https://bottosson.github.io/posts/oklab/#munsell-data

Everything in the article, from start to finish, seems to make perfect logical sense.

The problem is: to me, that perceptual colour wheel at the end does not appear to be perceptually uniform at all. The left hand side (yellow->green->blue) appears to vary far less than the right hand side (yellow->red->blue).

To me, the artist's one at the top looks most perceptually uniform, despite not being quite right from a physics perspective. The physics based one, with red/green/blue and yellow/cyan/magenta equally spaced is somewhere in between for me. This probably explains the myth of yellow/red/blue being the primary colours – they appear most perceptually different.

So why is that so if, by the sensible reasoning of the article, the wheel at the end should look most uniform? I don't know the answer to that. Maybe it's due to one of those later layers of processing mentioned? Maybe the non-uniformity is just baked into our brains for some evolutionary reason, to do with colours commonly encountered in the wild?

This was a good explanation - they simplify at points, but make it obvious they're doing so, which I'm fine with.

I would have liked to hear more about how the artists' color wheel came to be. It's a model, after all, and models don't have to be true as long as they're useful. The RYB color wheel is definitely not useful for setting colors in CSS, but as far as I know, the story goes something like this:

Before today's industrial processes, the colors of paint you could make or buy were those one could make from natural ingredients, for example cobalt for blue if I remember correctly. Artists and their suppliers noticed that if you had red, yellow and blue paint, all of which one could make from the materials available, then you could mix them to make any color you wanted (possibly with some white or black added to lighten/darken, or tint/shade if you're being pedantic). Further, you could make a lot of colors by mixing at most two of these colors, and it was purely the proportions of the mix that mattered, not the raw amounts.

So these got called "primary" colors, and it was natural to represent them in some way - to the engineer, a triangle seems like an obvious choice where the mix with 25% red and 75% blue lies 0.25 along the line from red to blue. Artists chose a circle instead, which is really the same model (mathematicians would say it's a topologically continous deformation from one to the other). At the time, the color wheel model turned out to be useful, and so it stuck around and got taught in art schools.

Magenta, Cyan and Yellow would have done just as well as starting points, but were harder if not impossible to get from the natural materials used at the time.

On the other hand, orange or brown as a starting color would not have got you the same mixing abilities as far as I'm aware. Brown is an interesting data point on its own because you can make it by taking orange and shading with a small amount of black, but you can also make it by taking orange and adding some blue (among many other ways), so now you're using all three primaries and fall off the color wheel. And that's before we talk about perceptual brightness and gamma correction and all that.

Red, Green and Blue on the other hand is a useful starting point for things that emit rather than absorb light (with similar but different caveats).

FWIW, I was exploring teaching color in K-2 using spectra. Here's[1] an old development snapshot, showing a color "wheel". Intent was a correct 3D perceptual color space (that snapshot isn't quite it), coupled with perceptual ("color") and physical ("light") spectra. Sort of the old art-school teaching from Munsell, but able to deal with materials and light. Absent an existing pedagogical perceptual color space (pedagogical: one without lots of "first thing you notice" features that are all bogus model artifacts), I kludged CAM16UCS with JzAzBz hue linearization and tweaks.

The author isn't wrong about current color instruction having dreadful content and outcomes. Ask first-tier non-astronomy physical-sciences graduate students what color the Sun is, and you commonly hear answers like "it doesn't have a color" and "it's rainbow color". Kids attempting to apply the models taught to their paints and pens and programs get... mixed results.

[1] https://imgur.com/a/zOtxQwe

> Ask any artist to explain how color works, and they’ll launch into a treatise about how the Three Primary Colors—red, blue, and yellow—form a color wheel

I don't think the author asked _any_ artists how color works before writing this line.

From the article:

//(Diagram) Josef Albers, Folder IV-1 Actually, the squares are exactly the same color! The surrounding context dictates the perceived color, on top of all that wavelength-physiology we just did.//

There's a difference between a color and a perceived color?

To add to the absurdity, I don't think you can argue that the example presents the same stimulus!

Maybe the author regards the term "color" as having a meaning in the sense of common-sense, except in this sense the common-sense sense of the sense has no common sense.

On what planet is blue the opposite of yellow? I was taught that blue is opposite of orange and yellow opposite of purple, which is symmetic.

I found the whole article a bit rigid and huffy. The standard color wheel and CMYK are just different ways of reducing color to basics from which other colors can be made. The watercolor set we used in elementary school is another, just less economical.

>If you are asked to imagine “a green with a bit of red,” nothing comes to mind.

Anyone else see an apple with that colour in their mind?

For a page about color, most of the images are sure washed out. Their white is visibly gray. I don't figure out why though. When you open up the images in a new tab they look fine, and I don't see opacity or a color transform in the CSS. It looks pretty bad though.

> In fact A and B are the same color (#787878), but you can’t see it even when you know this. To prove it to myself I had to open this picture in an image editor and actually move one square over another to see it was the same.

Looks like they're actually both #434343...

Goethe... his contemporary Schopenhauer has interesting notes on color theor, from an historical perspective

Wittgenstein also wrote some interesting notes on color. Both are more interesting than this blog post.

I'm not sure that I'd get "color wheels are wrong," from that, but it is a really good article on color perception and management, which is a huge topic.

(2011)

Some discussion then: https://news.ycombinator.com/item?id=2166494

Some interesting facts included, but a terrible presentation. The tendency to claim certain tools or visualizations are “wrong” for such a complex topic is a sign of a very incomplete understanding of the subject matter. We often learn about things in phases. Step 1: “Use this simple abstraction”, step 2 “actually that abstraction is ‘wrong’, here’s the real details”, step 3 “actually that abstraction was sort of right, in the appropriate context, with these caveats, as a tool, etc”.

Interesting. I didn't expect this kind of content from a blog I usually associate with giving startup advice.

As an artist I have a perspective on color from a pragmatic (as opposed to theoretical) point of view. Visual artists use materials like paints or inks applied to substrates. These all are exploiting subtractive color, primaries are complements of additive color, CMY vs. RGB.

The standard color wheel interleaves CMY with RGB, which sets up useful symmetries and complementary relationships. But of course, the color wheel is really a color cylinder, or set of cylinders when color properties other than hue are considered. Color saturation and value are crucial to artistic work, but when these additional dimensions are accounted for it's no longer the simple model the basic color wheel portrays.

Color classification systems emerged in the 20th century and over time evolved into the highly rigorous regimes codified by the International Color Consortium. Everybody is familiar with ICC profiles and their uses. I can't say I have a thorough understanding of them, I'd guess relatively few people do.

Artists don't have handy ICC profiles to guide creating a painting or print. They learn how to mix colors mainly by trial and error. However the paints and inks are precisely formulated by manufacturers who very much understand the science of color. This precision results in product constancy which makes life far easier for artists. With experience they gain an intuitive "fix" on particular colors and their locations on the extended color wheel.

To make it even more convoluted there are metallics, gold, silver, copper, etc. and the range of man-made iridescent and interference pigments. These "colors" aren't easy to merge with even multidimensional color "wheels" (or matrices). Nonetheless, people typically interpret these pigments as colors.

A subtle factor is opaque vs. transparent color effects. These aren't mutually exclusive, but in "pure" form are distinct. Opaque colors may be mixed to form an intermediate color, but at lower saturation and value. More than not this is how people are taught to manipulate colors often in accordance with the basic color wheel.

Transparent colors may be mixed like opaques, with somewhat similar results. However, layering transparent colors potentially produces extended effects. While both opaque mixing and layered color are subtractive, layering can combine in several ways producing multiple "virtual" colors modified by filtering effects of layers.

Any color wheel, or multidimensional matrix, is unlikely to accommodate the numerous ways colors combine or mix, or non-color colors. Artists use color wheels or more complex structures to gain pragmatic info about the colors they're using and predict results of manipulations. In any case, most real-world paints, inks, etc., don't behave like pure colors. A color wheel is only valuable as a pragmatically useful tool.

I wish they sold cmyk crayons and watercolors for children.

Once we can model the inner workings of the eyes, and optics of the physical world on "small" scale, accurately enough, it will be possible to collect lots of data on data. Consequently, an analysis of whether two invididuals of same skin color, medical profile (e.g. no cataplexy) and proportionally similar facial and body anatomy, that have near identical eyes, see colors differently in a laboratory environment. How can we make them to see colors differently by only making optical changes to the environment? Will we discover that species, age, sex or some mutation affects their "sensitivity" so that for some factors slight change can make colors seen differently?

The pattern: Control experimentally (IE slightly opposed to "statistically" where done afterwards) for the confluencing factors that seem obvious but you are uncertain are complete (IE can you "find / make two test-subjects significantly more identical with a reasonable amount of effort?"); choose something that can be observed in laboratory setting by the invididuals such that you can replicate the same environment for both of them; and now study this changing this thing the invididuals supposedly observe similarly at least in this and that environment by only changing the observed thing; seems to be more general, EG applicable to taste, smell. This may also actually not provide that much new scientific information.

What is the best neuroimaging definition for color that can be made? IE, a program that gives a probability of what seeing colors differently causes in the brain. I think we should, because our intuition is to teach children that animals also see colors at least partially, have the program be such that it is reasonably species agnostic. Surely the patterns one will need are out there...

Yet, it turns out to be more likely that we will end up with different definitions for all bugs and all big land animals or something similar. Then we would want to turn that into two different vocabularies. But we realize that how MUCH DOES EVERYTHING MAKE SENSE once we go that route and what we can see will no longer be what defines how we speak. The Urdu decided to base their vocabulary primary on this and that qualia research and the Latinists have developed a thousand times larger dictionary from the fact they prefer to highlight small taxonomical details even when the effect is small, and are now proposing based on neurological twin study of the differences between these two groups a new addition to the Esperanto language for describing how the musical experiences of speakers of both languages differ. Sixteen years later Urdu and Latin both have developed many new layers on top of their descriptions of the Esperanto research.

What new artifical languages will qualia research give birth to?

Will homoiconic programming languages have an application in cybernetics EG the program modifying it's internal structure based on how an implant detects your definitions and other patterns in thinking to change? And actually, what is the cybernetics of this: when a person given implant that reads and writes into their consciousness "need" to explicitly communicate that "No, I want THIS thing, THAT is my expression self, not what you the implant will read" to prevent "bad" loops?

Will these sorts of languages converge?

The language that "universally" describe "baseline" human experience are a sort of exit condition that prevents an endless recursive complexity from drowning us. I believe that for the sake of sanity once we have the cybernetics to ascend this should be kept as separate language aside the new languages that might araise, which could be launched from dead languages. This will probably occur naturally because the group that does not have cybernetics will retain their tongue; but imagine getting ripped of your cybernetics and due to your lack of knowledge of the baseline language being in a state of linguistic disability.

DNA modification may lead to the baselanguage rupturing, EG the babies who as I wrote this on 2024-07-12 were inside their mothers might have few who have been modified to gain a vision that also expands their gamut, or least alters it. This one could argue will not be a problem because we have always had color blind humans: but color blindless has always be a problem; and DNA guaranteeing perfect eyesight at cost of altered but not disable color perspection is NOT a maldaptive mutatition.

There is another safe anchor, more powerful than retaining baseline language, called computability theory. For the moment, no sapient creature on earth has even approached this, but eventually computational limits will force different languages arising from hopefully endless storms of the conscious experiencs to converge. If the irony that the AI zombie problem exists in philosophical literature is put aside, eventually with (if) multiple SAIs are sitting around Tellus and Milky Way talking to each other and modelling what is happening in the other's brain, they might reach a state where the rational action is to start unifying many languages speaking of the same topic instead of languages of qualia on top of each other. It might as well be that they are all versions of the same successful blueprint for being a SAI and there is not much difference in their "minds" because any would be an inefficiency.

Half of this is meaningless, but only because I had several things to express and don't stress about being wrong.

2011

Obligatory XKCD "Evolution of my understanding of color over time" https://xkcd.com/1882/

I'd say this article is around a half of the way down the panel