I agree that this blog post takes a sort of silly tone, and also that many artists are very sophisticated in their understanding and use of color.
He goes wrong in a few places, for instance in using quite confusing coloration of his diagrams, and he really shouldn’t put the labels R, G, B on cone responses: these are too evocative and are likely to confuse as much as they enlighten.
But to be fair, basic paint mixing models in artists’ heads often aren’t very good, and many artists would really benefit from studying human color vision more seriously. Explaining how vision works to visual artists, especially amateurs, is one of the more rewarding things I ever do: you can see these lightbulbs going off in their heads as a bunch of stuff that they know “intuitively” and experientially is given proper names and ordered into a comprehensible model.
> As for CMYK and RGB color spaces, these are simple but quite effective tools for dealing with color in a purely mechanical way.
I think these spaces have no place being taught to non-specialists. They are not intuitive to humans, being based on particular technology rather than on human vision. Any user interface which exposes these color spaces to non-experts is letting them down. Just as bad is using HSL or HSV or some similar trivial derivative space. Image editing software, in particular, is dramatically harder to use than it would be if the dimensions were at all relevant to vision.
> If you really need a color model more closely tied to how human visual physiology works, LAB was designed as a good approximation of that. Specialists sometimes use other, often proprietary systems.
Specialists looking for models related to vision mostly use open, publicly developed and specified models. At root, most of modern colorimetry is based on the CIE system, which has served very well since the 1930s. Recent fairly effective models that would be useful for software include CIECAM02 and IPT, which behave better in some ways than CIELAB, but are a bit more computationally expensive.
> If you need this level of knowledge, you got some book larnin' to do. No blog posts or comments on HN is going to help you too much.
That’s reasonable. Let’s start with some pointers to good sources!
For anyone who wants to really dive deep, I recommend http://www.handprint.com/LS/CVS/color.html (edit: Actually, I see that the blogger does link there. So that’s nice.)
Also, anyone looking for pointers to books, feel free to email me (address in profile), and I can point you in some direction based on your specific interests.
Artist's color mixing models aren't simple mechanical ratios. They use emergent properties of chemicals that just happen to be a particularly useful color. In fact, artist try to do as little color mixing as possible. Mixed pigments of the types artists use do work like the inks used in printing and the dyes used in RGB displays. Color wheels are guides and approximations. Learning a color theory that is more accurate is counter productive because you can't find pigments that behave that way. And if you ever get a chance to see a painting in the flesh and compare it to even a very good reproduction, you will see that the simple color model primaries used in printing and on screen reproduction are only loose approximations of the real colors.
Now what an artist does need to know is how color works perceptually. And the models used for this are pretty good. And remember the intended result is seldom "photo realism"; not that photos are particularly realistic. But that's another discussion altogether.
If you are talking about amateur artists, or beginners, then, sure, lecture away, but if you talk to someone who's been successfully painting for 40 years, my advice is to listen to him/her. They may or may not know much about the physiology of human vision, but don't be surprised if they do, but they certainly know the ins and outs of making you think you see what they want you to see.
The nice thing about standards are there are so many of them. I've spent more time than I cared to writing software to convert and compare between systems. They are a few lingua franca systems, but if you are doing serious, critical color work, your special cases are going to be the bulk of your job. Which only makes sense if you think about it. The sources I used are going to be about 30 years out of date, and the chemistries and optical systems no longer relevant.
Doing mechanical color reproduction is a specialty and few people need to know more than the basic RGB/CMY model and whatever specific systems that they need for their domain of activity. These days these models are very good and well tuned for their domains. Unless it's a brand new technology (Or NTSC; don't get me started!) This usually means working with a specific palette and color gamut.
And while a lot of people in the various industries feel that they are dealing with the real color models, and the rest mere approximations, there is no such thing. In the end it all breaks down and you are left with making specific measurements with specific types of equipment for use with specific use cases. And then you calibrate.
Color theory is very much a case of the map not being the territory.
> Learning a color theory that is more accurate is counter productive because you can't find pigments that behave that way
Learning theories can be tremendously helpful for crystalizing existing knowledge. Ultimately the only way to learn is to make visual art and look at it, a whole heck of a lot. But having theoretical models can help guide that thinking and looking, and can forestall a lot of confusion.
There are two kinds of models an artist needs to learn to deal with color:
(1) How the human eye sees color. This is so that texture and shape can be made to convey the desired effect of the picture. It’s helpful to learn about simultaneous contrast, adaptation effects, and even more obscure things like Ralph Evans’ (Kodak) theories about what he called “brilliance”. Learning this doesn’t tell you how to mix the paint, but it helps you figure out what results you should aim for.
(2) Color reproduction technology: the physical properties of paint / displays / printers / whatever.
Many people in the general public, including many artists, don’t properly understand the distinction between (1) and (2), and also take the radically oversimple concept of primary and secondary colors, etc. (whether it’s elementary school teachers with their red, yellow, blue, or programmers with their sRGB) as some kind of ultimate truth.
All of our models of color are approximations. Color is complex and contingent, depending on observer, lighting, viewing conditions, temporal effects, and so on.
My point is not that artists shouldn't learn about color and learn color theories. And I'm not saying that they shouldn't learn how human's see and process information. What I'm saying is that they already do that. But they are doing it with different models and techniques that are keyed to their domain and tools.
Sure, learn about RGB/CMY. It doesn't hurt. And printing technology, which is a different domain is very similar, though not the same in critical areas.
There's nothing magic or even very special about the methods and models you are talking about outside their domain of use.
If you are a designer who is working in print or onscreen display, you will learn to think in RGB/CMY(K). But there's nothing universal or special about that system compared to other systems. It's not more science compared to the systems artists use in their various working practices.
It's not more keyed to human physiology, in fact I'd argue that it is less intuitive and useful. I'll agree that it's not a numeric model and it's not a model designed for accuracy; and reproducibility. But it is a model tightly coupled with human perception, expression, and communication.
(As a designer working with digital imagery and renderings, I make major use of HSB style color models. I've written many special purpose tools for giving me effects I want, using all kinds of hair brained models, and then converting the results into RGB.)
To create a numeric model of how artists use color would be very difficult. If it weren't we'd be able to make cameras that made illustrations. And we can't and don't.
What an artist does is make an observation, and I mean this in both an abstract and practical sense. He or she then takes that observation and creates an abstraction of it, and then tries to make a work that communicates that abstraction. With the tools and mental model at hand.
There's a photo floating on Reddit right now of where a hair stylist has arranged hair clippings found on the floor in the shape of a dog. It's a wonderful piece of throwaway art. Now you could sit down and say, "If I take these hairs and arrange them with these percentages, I'll be able to get this range of colors and represent these textures ..."
But that's silly. The artist sees the hair, it reminds them of something, like a dog, and they spontaneously arrange them, making corrections and following intuition as they go.
Now if you are going to start an artistic tradition of arranging hair clippings for art, you might come up with some theories to model different effects. But I don't think you are going to be making numeric models. For one thing you never know what kinds of hairs you are going to find on the floor. Instead you are going to learn to "fake" it by using the relative and comparative artifacts of the human's visual and perceptual system. But it will almost certainly be intuitive and spontaneous. Lots of trial and error.
If this art form becomes a tradition, then special markets will develop in hair clippings, special dealers selling hair patches arranged nicely in special ways, maybe even a bit like RGB/CMY. (But don't push it.) Hair dyeing might become a specialty, but purists will insist on nature hair and natural hair colors. There will be much arguing about it and many bar fights.
And if you decide that there's a market in selling hair floor arrangements in mass quantities, then you might see something akin to what you see in modern printing or broadcast. Specialists will arise.
And most artists will say, "It's hard enough to make a pile of dog hair look like something interesting while worrying about all that. I'll make a good pile of hair and leave it up to the specialists to mass produce it. Maybe I'll use standard hair patches to make things easier, I do like it when the reproduction is well done. But I hate limiting my choices and opportunities, and besides the reproduced hair piles never look as good as the original anyway. After all, it's just a copy."
I think you’re misunderstanding me, because I mostly agree with you.
> There's nothing magic or even very special about the methods and models you are talking about outside their domain of use.
Learning about how cones & brains interpret light into the colors we perceive is definitely “special” compared to models designed around some particular physical medium like ink or dye or CRT displays, and the domain of use is “any time you want humans to look at colors”.
> [RGB/CMYK] is not more keyed to human physiology, in fact I'd argue that it is less intuitive and useful.
Sure. Which is why what I advocate teaching is the physiology and models aimed at approximating it (something with dimensions like the Munsell system or CIECAM02, etc.) ... which is absolutely more “intuitive” than RGB or CMYK or some paint mixing model, in the sense that it tracks perceived color attributes directly.
He goes wrong in a few places, for instance in using quite confusing coloration of his diagrams, and he really shouldn’t put the labels R, G, B on cone responses: these are too evocative and are likely to confuse as much as they enlighten.
But to be fair, basic paint mixing models in artists’ heads often aren’t very good, and many artists would really benefit from studying human color vision more seriously. Explaining how vision works to visual artists, especially amateurs, is one of the more rewarding things I ever do: you can see these lightbulbs going off in their heads as a bunch of stuff that they know “intuitively” and experientially is given proper names and ordered into a comprehensible model.
> As for CMYK and RGB color spaces, these are simple but quite effective tools for dealing with color in a purely mechanical way.
I think these spaces have no place being taught to non-specialists. They are not intuitive to humans, being based on particular technology rather than on human vision. Any user interface which exposes these color spaces to non-experts is letting them down. Just as bad is using HSL or HSV or some similar trivial derivative space. Image editing software, in particular, is dramatically harder to use than it would be if the dimensions were at all relevant to vision.
> If you really need a color model more closely tied to how human visual physiology works, LAB was designed as a good approximation of that. Specialists sometimes use other, often proprietary systems.
Specialists looking for models related to vision mostly use open, publicly developed and specified models. At root, most of modern colorimetry is based on the CIE system, which has served very well since the 1930s. Recent fairly effective models that would be useful for software include CIECAM02 and IPT, which behave better in some ways than CIELAB, but are a bit more computationally expensive.
> If you need this level of knowledge, you got some book larnin' to do. No blog posts or comments on HN is going to help you too much.
That’s reasonable. Let’s start with some pointers to good sources!
For this audience, Maureen Stone’s SIGGRAPH notes from 2001 are a nice first start, with lots of pointers to other good resrouces: http://graphics.stanford.edu/courses/cs448b-02-spring/04cdro...
For anyone who wants to really dive deep, I recommend http://www.handprint.com/LS/CVS/color.html (edit: Actually, I see that the blogger does link there. So that’s nice.)
Also, anyone looking for pointers to books, feel free to email me (address in profile), and I can point you in some direction based on your specific interests.