The browser demo (https://scrtwpns.com/mixbox/painter/) "runs two color mixing implementations in parallel: one based on Mixbox and the other that performs ordinary RGB mixing."
The issue is clicking between "Mixbox" & "Normal" in the app with unmixed paint shows two different starting colours which must make any comparison of mixed colours effectively void.
The starting colors are actually identical. Try painting a 100% opaque color pool and then switch between Mixbox and Normal. You'll see that the colors are the same. The difference is their behavior in thinner layers - when the brush opacity is lower.
When you spread dark pigments over a white canvas, they usually turn very radiant, saturated colors. Phthalo Blue turns from dark purplish blue into turquoise for example. RGB mixing doesn't handle those saturation gains and that's why a brush stroke made in the 'Normal' mode looks way more dull than the brush stroke made in 'Mixbox' mode with the identical starting color.
This is actually an important point of pigment mixing. The saturation gains and hue shifts when pigments get spread or mixed with white are beautiful. RGB linear mixing (the 'normal' mode) doesn't capture them.
This is correct, though I can certainly see how someone would come to the conclusion that starting colors are not identical, as default painting mode is less than full opacity, so the mixed color is generally a lot more saturated than plain sRGB.
And a clarification, I wouldn't call this RGB linear mixing, to me that's mixing in a physical light intensity space. Rather, it's alpha compositing in an sRGB color space, which is probably the most common mode. It's possible to verify this by setting transparency to exactly 50%, painting 50% black over white, and 50% white over solid black, verifying that the color is the same, and measuring that - it's very close to sRGB (128, 128, 128). Linear mixing would give approximately (188, 188, 188). Incidentally, this also shows that Mixbox doesn't retain neutral hue when mixing black and white - it gives something closer to (129, 127, 130), a slight reddish-purplish tone compared with neutral gray.
Colors are mixed internally in a four-component pigment space instead of RGB. To accelerate the conversion back to RGB for realtime painting, two 48 MB look-up tables are used.
The choice of these four primary pigments seems like a crucial artistic choice, because any color you choose on a traditional RGB/HSV picker is reinterpreted as a combination of these pigments. Seems like there might be an opportunity here to rethink the traditional color picker.
This algorithm is apparently implemented in a painting app called Rebelle. I haven't heard about it before, but I think I'll buy a license because it looks awesome:
I saw the same thing. I watched the video and they seemed to choose the 4 primary pigments to be close to CMYK, but I was also surprised they didn't make a lookup table for other CMYK pigment alternatives. Seemed like it wouldn't be crazy to do and gives another nice artistic knob to turn. And if it didn't affect much, that would also be convincing that they weren't too sensitive to the 4 pigment selection which is also nice
(Although, academic paper cherry picking aside, I was pleasantly surprised how visually close some of their examples were to the read world, for different pigments)
what the average eye sees from a given pigment (and light source) is not what Everybody sees; and we have a hard time discussing those fine gradations of perception.
My wife sees enough infrared to see hot air circulation in indoor light; it makes her art interesting because she literally sees the world in a different way than most others and can express some of that in a way that we can see... it looks trippy, mostly :)
She describes the diffraction of light when hot air moves; but that light isnt visible to most and its not a color she can describe well. "Sees infrared" may not be accurate, but its the best guess we've got. Her colors are shifted and slightly wider gamut than others too, as far as we can tell. She sees some of the "UV flower patterns" but not as a stark contrast, just more pattern and depth in a flower petal than I (or a camera) can see.
We have a big propane fired ceramic element heater; it lights the room for her more than it does for me. She can see the air dancing over it clearly by its own light, i have to smoke near it and have a lamp to see the same thing.
Sounds like she sees the refraction of light through the different densities of air, aka a mirage. Artists are usually more perceptive of how light contributes to what we see.
This is a great paper, will likely move the state of the art forward. I've been aware of Kubelka-Munk representations of painting surfaces since [1], which used a six-dimensional space (scattering and absorption for 3 wavelength bands, not really spectral). Doing it all in RGB makes it way more practical.
I'm fiddling with an alternative approach based on IPT-like color spaces, which won't need the 10MB lookup table, and will have a tunable parameter for paint opacity (between watercolors and pastels). So far I have promising but not compelling results. If people are working on, say, a free software painting application, get in touch.
15 comments
[ 4.5 ms ] story [ 45.5 ms ] threadThe issue is clicking between "Mixbox" & "Normal" in the app with unmixed paint shows two different starting colours which must make any comparison of mixed colours effectively void.
This is actually an important point of pigment mixing. The saturation gains and hue shifts when pigments get spread or mixed with white are beautiful. RGB linear mixing (the 'normal' mode) doesn't capture them.
And a clarification, I wouldn't call this RGB linear mixing, to me that's mixing in a physical light intensity space. Rather, it's alpha compositing in an sRGB color space, which is probably the most common mode. It's possible to verify this by setting transparency to exactly 50%, painting 50% black over white, and 50% white over solid black, verifying that the color is the same, and measuring that - it's very close to sRGB (128, 128, 128). Linear mixing would give approximately (188, 188, 188). Incidentally, this also shows that Mixbox doesn't retain neutral hue when mixing black and white - it gives something closer to (129, 127, 130), a slight reddish-purplish tone compared with neutral gray.
Colors are mixed internally in a four-component pigment space instead of RGB. To accelerate the conversion back to RGB for realtime painting, two 48 MB look-up tables are used.
The choice of these four primary pigments seems like a crucial artistic choice, because any color you choose on a traditional RGB/HSV picker is reinterpreted as a combination of these pigments. Seems like there might be an opportunity here to rethink the traditional color picker.
This algorithm is apparently implemented in a painting app called Rebelle. I haven't heard about it before, but I think I'll buy a license because it looks awesome:
https://www.escapemotions.com/products/rebelle/about
(Although, academic paper cherry picking aside, I was pleasantly surprised how visually close some of their examples were to the read world, for different pigments)
My wife sees enough infrared to see hot air circulation in indoor light; it makes her art interesting because she literally sees the world in a different way than most others and can express some of that in a way that we can see... it looks trippy, mostly :)
Infrared should not be very prominent even if you could see it. It might be a slightly red tone I would argue.
Hot air doesn't radiate much energy.
You don't mean by any chance the fliring effect when warm air leaves a window?
We have a big propane fired ceramic element heater; it lights the room for her more than it does for me. She can see the air dancing over it clearly by its own light, i have to smoke near it and have a lamp to see the same thing.
I'm fiddling with an alternative approach based on IPT-like color spaces, which won't need the 10MB lookup table, and will have a tunable parameter for paint opacity (between watercolors and pastels). So far I have promising but not compelling results. If people are working on, say, a free software painting application, get in touch.
[1]: https://www.levien.com/gimp/wetdream.html