This is beautiful work! First, I thought this would involve using PCA to go from 3D to 2D, which would result in an easier selector to use, but at the expense of representing every person.
Then, I thought this would stop at using the U-space vectors (the ones that form the basis of the PCA image) and their corresponding ellipse to form our color space, but no, the function fitting is a very slick idea, even if it was executed by hand.
Lastly, I love the presentation of sampling from different r values. Whether you sample from a fixed r value or a range of r values, I bet this has great applications in game design or animation.
I still don't quite get the manual data labeling process at the beginning? It seems like it would encode some bias, but the consistency of the first point cloud and the results certainly speak for themselves.
If you look at my doctoral thesis, in fact skin colour can be parametrised using melanin and haemoglobin concentration; which it means it is indeed (roughly) two dimensional. This allows you to invert the colour of a pixel and identify someone’s chromophore concentration. This is based on previous work of others cited in the thesis.
Oh goody, an expert in a question I've always wanted to ask.
Once upon a time, I encountered a sort of interactive museum book called "Earthsearch". One of its "exhibits" was a color wheel, made of two spinnable pieces of tinted acetate, that purported to be able to generate any skin tone. I suppose this is three degrees of freedom, since for any position of the two wheels you could access a range of colors around the edge - but the printed page underneath was simply a grayscale gradient, so one degree of freedom was simply luminance.
The book was released in 1994, and was very confident. It even went as far as to encourage the reader to see a doctor, should their skin tone not be achievable! What color model were they using?
That’s sounds very interesting. I’m not sure. In the thesis, I generated spectral curves and transformed them using the light and camera sensitivity curves. So, in theory it could be used in any color space.
But I can’t vouch for the accuracy of the book, but I suspect it was an approximation.
The haemoglobin and melanin are roughly in separate layers of the skin. There is an approximation known as Beer-Lambert that models the skin as two semi-transparent layers. This ignores subsurface scattering, but is roughly correct.
So, in theory two acetate sheets might replicate this. That’s my intuitive opinion. It has been many years since I did this work.
I loved that book! Was it that book or its sister that had an image of the pole and said "Spin the Earth the correct direction"? That and "How Things Work" kicked so much ass
I've heard that the so-called "yellowish" skin color that non-tanned East Asians tend to have, is due to relative abundance of subcutaneous fat. If that's true, is that captured by the haemoglobin dimension or is it a yet another degree of freedom?
We used the approximation that the hypodermis was a perfectly white reflector; so the thickness wouldn’t matter. The collagen fibrills scatter light in such a way to create a yellower tint. So a person’s skin might appear this way in the absence of melanin and haemoglobin. I’m not an expert on population statistics, and so cannot give an opinion on specific skin types. Everything was based on ranges.
There is a separate effect of increased bilirubin content, that would appear if someone was ill.
This brings up an interesting topic of atypical compounds that can visibly affect skin coloring -- whether temporary or more long-lived (such as elevated methemoglobin levels in the Blue Fugates of Kentucky).
It's fascinating to learn that melanin and haemoglobin are the two primary factors (and cover the vast majority of cases), but I'm curious how many other illnesses / disorders can produce compounds with visible effects.
Thank you for your posts on this thread -- I appreciate you knowledge-sharing on this!
In theory, you can construct a skin model of any condition you want. It would be a case of knowing the concentration, scattering profiles and reflectance spectra of any chromophores. For the thesis, I wrote a Monte Carlo skin simulator using OpenCL, when it has just come out. You could plug anything in there.
One thing about East Asian yellow tint in skin is that there really isn't such a thing. Asian base skin color is same as Europeans, if slightly more reddish or slightly more fatty or whatnot, but more or less within margins of error. It's largely an invention from late Age of Discovery era to feed to us-them mentality, with perhaps some effect of malnourishment and sunlight some deep southeast guys was living with at the time of their first contacts.
Not an expert, but when I was learning 3d modeling, there was a good chapter on subsurface scattering which affect the human skin rendering. It’s what give it liveness instead of looking like a doll.
All skin is slightly translucent, and color/tones forms a gradient across the surface of the skin. Your upper cheeks and forehead are usually slightly reddish, and the chin/neck are slightly bluish.
People only turn yellow if they have kidney or liver damage. However, some distinctive features like Lucy Liu's natural freckles are tough to draw right.
We looked around a lot for a procedural texture/shader to help make the project characters look less sickly, and please also consider supporting the persons work if you are able:
It is a classic artist starting off point to first drop a red/blue gradient as a base mixing layer, and add skin details/masks for each additional texture layer/feature.
The "The Ultimate Blender 3D Sculpting Course" covers several sculpting techniques, and while this course has a number of challenges it does cover several basics. Probably wise to wait for a sale event if interested.
> People only turn yellow if they have kidney or liver damage.
This is not true. Roughly 5% of people have Gilbert's Syndrome which is genetic, metabolic abnormality where the bilirubin from dead blood cells are conjugated slower than the baseline patient leading to yellowing of the skin. It is not due to liver damage or disease and is a benign condition. Not to mention there is carotenemia from eating too many carrots or squash -- again not kidney/liver damage related.
Thank you! I call out the manual labeling bias and also that a future work improvement would be to be scientific about that labeling process (multiple labelers, expert labelers), and then to feed that into the equation generation steps. Part of where the "good enough" naming comes from - I was hoping that using my art brain and keeping a bunch of reference images in mind (shout out to the Humanae project) would maybe offset the bias a bit. But 100% right on the bias being there.
It's always entertaining to try to understand these things from first principles. Color and skin color in particular are very complex things to measure and model. It's not just a physical quantity, it's also a matter of human perception based on lighting and a host of factors.
It seems like some work was done to research existing approaches, but I don't see any reference to Pantone Skin Tones [0]. Without intending to, I've worked on 3 different projects which involved measuring skin tones for makeup applications. The Pantone range is not 100% perfect, but it's pretty well done and easy to understand. There are two axes: a light/dark one, and red-yellow one [1].
There's also the Monk Skin Tone Scale by Dr. Monk and Google (available under a CC attribution 4.0 international license). It was developed for machine learning, so it's not exactly the same use case, but it provides 10 colors in a variety of formats, including hexadecimal. https://skintone.google/get-started
Pantone is not a color space. They are a supplier of physical color samples. Most pigments can be represented in a color space, but that representation differs based on illumination so it not unique. Some Pantone pigments, fluorescent colors in particular, cannot be represented by a color space at all!
Yup. This is a rabbit hole like time. Learn a little bit about it, know there's a staggering amount more and then any time you actually need it, check that you've agreed on as much detail as you needed for the task and not less.
The person inviting you to a picnic "next week" might need to know about Time Zones if they live some distance away. The requirement to update the CSS on the web site to match a new company logo might need to know about Colour Spaces. But that picnic invite doesn't care about Calendar systems or Leap Seconds, and the logo mandate doesn't care about Sub-surface scattering or a Quantum theory of light.
> What colors are we? The short answer is maybe something like “brown” and the long answer is very, very long.
Pete Shirley (author of Ray Tracing In One Weekend) is also a part time color scientist, and he clued me into the fact that if you crank the saturation to 100% of pictures of people of any race, the skin color comes out orange. I think I heard that some face detectors may work on this principle, but I don’t remember where I heard that.
Yes, true! Though cranking saturation doesn’t brighten colors (much… at least it’s not supposed to), and maybe interestingly, light brown isn’t quite orange. There might some asymmetry in color names.
So it might not be that surprising that dark skin people turn orange when sat is cranked, but perhaps slightly surprising that light skin people do.
Anyway, I suspect it’s computationally easier to turn everyone orange than brown. Plus, given that brown skin is already a category, orange could be useful terminology/category.
Nice. Most of these colors look right to me, but I'm also seeing green, blue and purple? If you're curating colors from the space manually then it's not a problem.
"Only realistic colors" would be a different problem. And I'm not sure if there's a way to solve that problem without excluding some real colors too.
(Maybe if the colour selector was a different shape? I don't know much about color spaces.)
Yes absolutely right on - I mention much deeper in the page that in a picker UI you can use a bigger range, but that in a generation context you might want to be a bit more selective. I go into it a bit more when I explain what R^2 is.
"not sure if there's a way to solve that problem without excluding some real colors too" - 100% right on that too. A big part of this work was to define a way to tweak/improve the equations in order to approach a solution to these issues - part of my hope in sharing is to get more eyes and brains thinking about this and proposing improvements!
Reducing the R^2 value to ~1.5 excludes skin colors that are extremely blue, green and purple. That seems to be a happy medium between including real outliers and excluding colors like green.
If you drag the bottom slider all the way to the right you only have plausible skin tones. Drag it all the way to the right and you can match skin tones for orcs, Na'vi and djinn
Love it. I especially thought the introspective “aside” section on related resources was great, and I wish more people would show this kind of reflection.
Thanks! I hope to send lots of love those ways, they're basically all things that I found outside of this work that I liked enough to remember (I think Humanae was the one that I found while doing this).
I definitely don't speak for all non-white people when I say this, but personally I find people obsessing over it disturbing/offensive independent of whether their intentions are racist or anti-racist.
I did find modeling/data-driven perspective cool though
I also agree. Yes, humans need to be reminded that not all humans are just like their immediate group, but that goes for all humans, not just the white ones. And I certainly don't get accusations of racism that come along with it. But then I don't really get the needing to be represented thing either. Was I represented by Mario???
I think it’s worthwhile to note this is under a light with a color temperature similar to sunlight.
When you’re actually painting, skin can really be any color. Blue is a perfectly valid (and fairly common) skin color - just look at someone in the moonlight. It takes two to tango, and the other dancer is the light source.
Interesting though that the 1.5 setting (at least on my older iPhone) doesn’t really ever seem to cover the range of skin tones of indigenous Australians (“aboriginal” and Torres Strait Islanders), who apparently get their skin tones from a somewhat different expression of genes than other peoples with darker skin.
The 2.0 “outliers” cover those skin tones.
Masai and some Somali skin tones, similarly, look better represented by the outliers.
The full range of possible human skin tones is perhaps a bit of a challenge for technical colour representation, partly because diffuse reflectance seems to vary so much and that is hard to approximate on screen (e.g. as any website designer could attest, the difference between representing silver and grey is the use of highlights), though the representation of skin tones is a lot better than the early colour film era.
Yes, 2.0 is the value I tuned the equations around - but I think there's definitely room for improvement in terms of squashing the space along the fair/deep axis in order to get better results at lower R^2.
The issue I ran into was what happens when the space moves beyond RGB, but I think the fix is to just add clamps to the end R, G, B values.
This is super cool! I have always loved projects that try to categorize color space. Usually my work has gone in the other direction - isGray, isCloseToWhite, isPastel - but it's the same insofar as it's defining shapes like your matplot visual.
Wish I had this on hand like this three or four years ago when I was working on VR avatar software - we had hard-coded skin tone options at the time - would have been a cool addition!
In sharing this around I've found more people than I thought who've done things to semantically chop up the color space - definitely a more interesting area than I first thought of when working on my narrow use case. Feels like it'd be fun to play with something that has a whole bunch of different regions of color space labeled - like "here's where IsPastel overlaps with IsGreen".
But absolutely the hope is that this is a tool that stays accessible to people for whatever they're building and everyone can slow down on reinventing the wheel here for every character creator/digital art application.
The U and V axes seem roughly correspond to the tint/temperature (green-magenta/blue-amber) axes that you can find in a white balance tool in photo editors and also as an in-camera setting.
This seems to do XYZ->RGB without any gamma correction. XYZ is linear, as is the transform to RGB, so we usually convert linear RGB to (gamma-encoded) sRGB at the end.
The JS code seems to output linear RGB directly to the webpage (missing gamma). I guess the difficulty with "adding gamma" to it is this: if the training data were also gamma encoded, then it might need to be refit with linear values.
I read the “limitations” section’s list of things that affect perceived skin tone and it didn’t mention lighting! This is pretty important as the available light determines the shades that can be perceived.
Skin tones are particularly difficult to pin down, besides the pigmentation there are two effects at play: 1) sub surface scattering (light goes thru and bounces under the skin a bit) and 2) some infra-red emission (the responsible for making some people appear magenta on digital photography)
So even when you get the palette right, to paint it like a believable material in a convincing way is still super hard!
Could you explain the first step, "Manually Labeling Colors ? What did you manually label? I didn't quite follow.
From the way the article is written, it sounds like you presented yourself with a bunch of faces of completely random colours and then manually sorted them into two categories, "plausible skin colour" and "implausible skin colour".
But I feel like I must be misunderstanding, because I don't see why you would do that. Why wouldn't you gather photographs of actual faces and sample colours directly from the photos?
I think the ideal would be to use this picker to select a few tones, and then gradient map those across different parts of the body according to a grayscale map. Then you could fine-tune things like palm color, cheeks, facial tone, etc. individually.
You mention a bit about the latent social forces embedded within technological development and how that has related to skin tones and the technologies of color reproduction, so I figured I would share some references:
The Kodak “Shirley cards” and test shots are of course famous in the media theory world for how they baked notions of race and “standard” or “normal” conditions into the medium of photography. Not dissimilar from the various discussions around facial recognition from approximately 10 yr ago.
Very neat project. Could I use this code to generate skin tones in my own project? Or more meant to just be educational/inspirational? I guess I'm wondering if you've considered adding a license.
Good catch, added MIT license - feel free to use it! On top of that, my understanding is that math/equations are not subject to copyright, so the core part of this work (the equations/ideas) I think legally has 0 limitations - even if there were a restrictive license. But IANAL so might have some misconceptions.
ymmv, as some paid plugins actually speed up the funky blender workflow (the auto-Bake plugin by the same author is important for procedural generated textures.)
I'm using an iPhone Safari plugin called Noir to force darkmode and your tool is making white the absence of color. In other words, I think white is just an alpha channel.
Nice. This is relatively old work (2020) but Pinterest had/has skin tone features, for example if you're searching for cosmetic products. Worth a check if anything might be useful for your work.
https://medium.com/pinterest-engineering/powering-inclusive-...
On the movie Dolemite by Eddy Murphy, they go over the fact that black people absorb light and white people reflect, and how that changes the motion picture shooting process totally. Also how one size does not fit all. Cool project! I am not much into the humanities part of the project. For me this is as natural as having a localization framework, just in this case for skin tones.
Feature request: drop-down list of ethnicities, where the selection turns on specific guard rails in the parameter space (or retains full movement of the controls, but scales the values). Or is that too racist for 2026?
I really like the idea of treating a color space as a task-specific optimization rather than something universal. It makes me wonder if we're underusing domain-specific color spaces in procedural generation in general.
Setting R² to various large values like 50, 500 and 5000 then moving the slider around seems to give a decent and simple visual representation of what's happening.
Not my domain though I followed the data science enough to nod as if I really followed it,
but this is just lovely and inspirational work. A unique marriage of hacking in the celebratory sense, thoughtfulness, and meaningful contribution. Outstanding.
Skin is so interesting as a surface to think about how light passes through and bounces off of.
There is at least one article published by a technical artist at Pixar who describes modeling a skull within a human scan so that subsurface lighting is more accurately rendered.
I suspect you could combine this with dr_scully’s comment and GolDDranks’ to reach the nuance involved in long tail realism details.
There's a lot of really good research papers that touch on this area from what I saw - would definitely be cool to see if that extra nuance could be added without dramatically increasing the math complexity; in its current state, the code could be used in any programming language without a library.
yeah, skin tone variations often feels it's about limiting lighter tone options while adding luminosity options for the exact same brown color. There seem to be B channel differences across darker colored people, and I wouldn't be sure if they are better represented either.
This is beautiful work! First, I thought this would involve using PCA to go from 3D to 2D, which would result in an easier selector to use, but at the expense of representing every person.
Then, I thought this would stop at using the U-space vectors (the ones that form the basis of the PCA image) and their corresponding ellipse to form our color space, but no, the function fitting is a very slick idea, even if it was executed by hand.
Lastly, I love the presentation of sampling from different r values. Whether you sample from a fixed r value or a range of r values, I bet this has great applications in game design or animation.
I still don't quite get the manual data labeling process at the beginning? It seems like it would encode some bias, but the consistency of the first point cloud and the results certainly speak for themselves.
If you look at my doctoral thesis, in fact skin colour can be parametrised using melanin and haemoglobin concentration; which it means it is indeed (roughly) two dimensional. This allows you to invert the colour of a pixel and identify someone’s chromophore concentration. This is based on previous work of others cited in the thesis.
https://discovery.ucl.ac.uk/id/eprint/10093345/
Oh goody, an expert in a question I've always wanted to ask.
Once upon a time, I encountered a sort of interactive museum book called "Earthsearch". One of its "exhibits" was a color wheel, made of two spinnable pieces of tinted acetate, that purported to be able to generate any skin tone. I suppose this is three degrees of freedom, since for any position of the two wheels you could access a range of colors around the edge - but the printed page underneath was simply a grayscale gradient, so one degree of freedom was simply luminance.
The book was released in 1994, and was very confident. It even went as far as to encourage the reader to see a doctor, should their skin tone not be achievable! What color model were they using?
That’s sounds very interesting. I’m not sure. In the thesis, I generated spectral curves and transformed them using the light and camera sensitivity curves. So, in theory it could be used in any color space.
But I can’t vouch for the accuracy of the book, but I suspect it was an approximation.
The haemoglobin and melanin are roughly in separate layers of the skin. There is an approximation known as Beer-Lambert that models the skin as two semi-transparent layers. This ignores subsurface scattering, but is roughly correct.
So, in theory two acetate sheets might replicate this. That’s my intuitive opinion. It has been many years since I did this work.
Oh man, that's a blast from the past, I think I had that book! Was that the one with an aluminum cover?
I loved that book! Was it that book or its sister that had an image of the pole and said "Spin the Earth the correct direction"? That and "How Things Work" kicked so much ass
That's it! They (rather dubiously) claimed it came from a recycled Russian coke can, if I recall.
I had that one too. I remember I used to crunch on the rice from the bag it came with (I was like 8).
I've heard that the so-called "yellowish" skin color that non-tanned East Asians tend to have, is due to relative abundance of subcutaneous fat. If that's true, is that captured by the haemoglobin dimension or is it a yet another degree of freedom?
We used the approximation that the hypodermis was a perfectly white reflector; so the thickness wouldn’t matter. The collagen fibrills scatter light in such a way to create a yellower tint. So a person’s skin might appear this way in the absence of melanin and haemoglobin. I’m not an expert on population statistics, and so cannot give an opinion on specific skin types. Everything was based on ranges.
There is a separate effect of increased bilirubin content, that would appear if someone was ill.
This brings up an interesting topic of atypical compounds that can visibly affect skin coloring -- whether temporary or more long-lived (such as elevated methemoglobin levels in the Blue Fugates of Kentucky).
It's fascinating to learn that melanin and haemoglobin are the two primary factors (and cover the vast majority of cases), but I'm curious how many other illnesses / disorders can produce compounds with visible effects.
Thank you for your posts on this thread -- I appreciate you knowledge-sharing on this!
In theory, you can construct a skin model of any condition you want. It would be a case of knowing the concentration, scattering profiles and reflectance spectra of any chromophores. For the thesis, I wrote a Monte Carlo skin simulator using OpenCL, when it has just come out. You could plug anything in there.
One thing about East Asian yellow tint in skin is that there really isn't such a thing. Asian base skin color is same as Europeans, if slightly more reddish or slightly more fatty or whatnot, but more or less within margins of error. It's largely an invention from late Age of Discovery era to feed to us-them mentality, with perhaps some effect of malnourishment and sunlight some deep southeast guys was living with at the time of their first contacts.
Not an expert, but when I was learning 3d modeling, there was a good chapter on subsurface scattering which affect the human skin rendering. It’s what give it liveness instead of looking like a doll.
All skin is slightly translucent, and color/tones forms a gradient across the surface of the skin. Your upper cheeks and forehead are usually slightly reddish, and the chin/neck are slightly bluish.
People only turn yellow if they have kidney or liver damage. However, some distinctive features like Lucy Liu's natural freckles are tough to draw right.
We looked around a lot for a procedural texture/shader to help make the project characters look less sickly, and please also consider supporting the persons work if you are able:
https://kuhantilope.gumroad.com/l/ultimate-skin
Best of luck =3
I didn't know chin/neck are slightly bluish, can we make it more explicit without side effects?
It is a classic artist starting off point to first drop a red/blue gradient as a base mixing layer, and add skin details/masks for each additional texture layer/feature.
The "The Ultimate Blender 3D Sculpting Course" covers several sculpting techniques, and while this course has a number of challenges it does cover several basics. Probably wise to wait for a sale event if interested.
https://www.udemy.com/course/blender-3d-sculpting-course/
Best of luck =3
> People only turn yellow if they have kidney or liver damage.
This is not true. Roughly 5% of people have Gilbert's Syndrome which is genetic, metabolic abnormality where the bilirubin from dead blood cells are conjugated slower than the baseline patient leading to yellowing of the skin. It is not due to liver damage or disease and is a benign condition. Not to mention there is carotenemia from eating too many carrots or squash -- again not kidney/liver damage related.
Gilbert's is a liver disease...
https://my.clevelandclinic.org/health/diseases/17661-gilbert...
Not in dr_scully’s reply is that his thesis says haemoglobin is not static. Blood flow to the skin can vary over time.
Thank you! I call out the manual labeling bias and also that a future work improvement would be to be scientific about that labeling process (multiple labelers, expert labelers), and then to feed that into the equation generation steps. Part of where the "good enough" naming comes from - I was hoping that using my art brain and keeping a bunch of reference images in mind (shout out to the Humanae project) would maybe offset the bias a bit. But 100% right on the bias being there.
It's always entertaining to try to understand these things from first principles. Color and skin color in particular are very complex things to measure and model. It's not just a physical quantity, it's also a matter of human perception based on lighting and a host of factors.
It seems like some work was done to research existing approaches, but I don't see any reference to Pantone Skin Tones [0]. Without intending to, I've worked on 3 different projects which involved measuring skin tones for makeup applications. The Pantone range is not 100% perfect, but it's pretty well done and easy to understand. There are two axes: a light/dark one, and red-yellow one [1].
[0] https://www.pantone.com/na/en-us/skintone
[1] https://cosmeticsbusiness.com/pantone-skintone-guide-80945
There's also the Monk Skin Tone Scale by Dr. Monk and Google (available under a CC attribution 4.0 international license). It was developed for machine learning, so it's not exactly the same use case, but it provides 10 colors in a variety of formats, including hexadecimal. https://skintone.google/get-started
well its not a free as in beer color space
Pantone is not a color space. They are a supplier of physical color samples. Most pigments can be represented in a color space, but that representation differs based on illumination so it not unique. Some Pantone pigments, fluorescent colors in particular, cannot be represented by a color space at all!
Yup. This is a rabbit hole like time. Learn a little bit about it, know there's a staggering amount more and then any time you actually need it, check that you've agreed on as much detail as you needed for the task and not less.
The person inviting you to a picnic "next week" might need to know about Time Zones if they live some distance away. The requirement to update the CSS on the web site to match a new company logo might need to know about Colour Spaces. But that picnic invite doesn't care about Calendar systems or Leap Seconds, and the logo mandate doesn't care about Sub-surface scattering or a Quantum theory of light.
Love it!
I was looking at this a little while ago, and used some of The Pudding's data on makeup/foundation shades (https://pudding.cool/2018/06/makeup-shades/) and plotted it into the Oklab colorspace (https://observablehq.com/@55th/foundation-shades). The shades form themselves into that same crescent as seen in the article.
Woah I LOVE both of these! When I get a chance these are 100% getting linked from the writeup.
> What colors are we? The short answer is maybe something like “brown” and the long answer is very, very long.
Pete Shirley (author of Ray Tracing In One Weekend) is also a part time color scientist, and he clued me into the fact that if you crank the saturation to 100% of pictures of people of any race, the skin color comes out orange. I think I heard that some face detectors may work on this principle, but I don’t remember where I heard that.
And brown is just orange... with context.
https://www.youtube.com/watch?v=wh4aWZRtTwU
and brown is just dark orange.
Yes, true! Though cranking saturation doesn’t brighten colors (much… at least it’s not supposed to), and maybe interestingly, light brown isn’t quite orange. There might some asymmetry in color names.
So it might not be that surprising that dark skin people turn orange when sat is cranked, but perhaps slightly surprising that light skin people do.
Anyway, I suspect it’s computationally easier to turn everyone orange than brown. Plus, given that brown skin is already a category, orange could be useful terminology/category.
And orange is just a yellow red.
Or a red yellow
Nice. Most of these colors look right to me, but I'm also seeing green, blue and purple? If you're curating colors from the space manually then it's not a problem.
"Only realistic colors" would be a different problem. And I'm not sure if there's a way to solve that problem without excluding some real colors too.
(Maybe if the colour selector was a different shape? I don't know much about color spaces.)
Yes absolutely right on - I mention much deeper in the page that in a picker UI you can use a bigger range, but that in a generation context you might want to be a bit more selective. I go into it a bit more when I explain what R^2 is.
"not sure if there's a way to solve that problem without excluding some real colors too" - 100% right on that too. A big part of this work was to define a way to tweak/improve the equations in order to approach a solution to these issues - part of my hope in sharing is to get more eyes and brains thinking about this and proposing improvements!
I would have "show sphere" on by default--this kind of threw me at first, too.
This is a really nice project. Great work and a great write-up. Thanks for sharing.
> blue
You'd be surprised: https://en.wikipedia.org/wiki/Argyria
Reducing the R^2 value to ~1.5 excludes skin colors that are extremely blue, green and purple. That seems to be a happy medium between including real outliers and excluding colors like green.
If you drag the bottom slider all the way to the right you only have plausible skin tones. Drag it all the way to the right and you can match skin tones for orcs, Na'vi and djinn
Love it. I especially thought the introspective “aside” section on related resources was great, and I wish more people would show this kind of reflection.
Thanks! I hope to send lots of love those ways, they're basically all things that I found outside of this work that I liked enough to remember (I think Humanae was the one that I found while doing this).
I definitely don't speak for all non-white people when I say this, but personally I find people obsessing over it disturbing/offensive independent of whether their intentions are racist or anti-racist. I did find modeling/data-driven perspective cool though
I also agree. Yes, humans need to be reminded that not all humans are just like their immediate group, but that goes for all humans, not just the white ones. And I certainly don't get accusations of racism that come along with it. But then I don't really get the needing to be represented thing either. Was I represented by Mario???
I think it’s worthwhile to note this is under a light with a color temperature similar to sunlight.
When you’re actually painting, skin can really be any color. Blue is a perfectly valid (and fairly common) skin color - just look at someone in the moonlight. It takes two to tango, and the other dancer is the light source.
Interesting though that the 1.5 setting (at least on my older iPhone) doesn’t really ever seem to cover the range of skin tones of indigenous Australians (“aboriginal” and Torres Strait Islanders), who apparently get their skin tones from a somewhat different expression of genes than other peoples with darker skin.
The 2.0 “outliers” cover those skin tones.
Masai and some Somali skin tones, similarly, look better represented by the outliers.
The full range of possible human skin tones is perhaps a bit of a challenge for technical colour representation, partly because diffuse reflectance seems to vary so much and that is hard to approximate on screen (e.g. as any website designer could attest, the difference between representing silver and grey is the use of highlights), though the representation of skin tones is a lot better than the early colour film era.
Yes, 2.0 is the value I tuned the equations around - but I think there's definitely room for improvement in terms of squashing the space along the fair/deep axis in order to get better results at lower R^2.
The issue I ran into was what happens when the space moves beyond RGB, but I think the fix is to just add clamps to the end R, G, B values.
This is super cool! I have always loved projects that try to categorize color space. Usually my work has gone in the other direction - isGray, isCloseToWhite, isPastel - but it's the same insofar as it's defining shapes like your matplot visual.
Wish I had this on hand like this three or four years ago when I was working on VR avatar software - we had hard-coded skin tone options at the time - would have been a cool addition!
In sharing this around I've found more people than I thought who've done things to semantically chop up the color space - definitely a more interesting area than I first thought of when working on my narrow use case. Feels like it'd be fun to play with something that has a whole bunch of different regions of color space labeled - like "here's where IsPastel overlaps with IsGreen".
But absolutely the hope is that this is a tool that stays accessible to people for whatever they're building and everyone can slow down on reinventing the wheel here for every character creator/digital art application.
The U and V axes seem roughly correspond to the tint/temperature (green-magenta/blue-amber) axes that you can find in a white balance tool in photo editors and also as an in-camera setting.
Nb. I'd advise against calling the PCA'd space XYZ because that's a well known standard color space: https://en.wikipedia.org/wiki/CIE_1931_color_space#CIE_XYZ_c...
This seems to do XYZ->RGB without any gamma correction. XYZ is linear, as is the transform to RGB, so we usually convert linear RGB to (gamma-encoded) sRGB at the end.
The JS code seems to output linear RGB directly to the webpage (missing gamma). I guess the difficulty with "adding gamma" to it is this: if the training data were also gamma encoded, then it might need to be refit with linear values.
I read the “limitations” section’s list of things that affect perceived skin tone and it didn’t mention lighting! This is pretty important as the available light determines the shades that can be perceived.
Skin tones are particularly difficult to pin down, besides the pigmentation there are two effects at play: 1) sub surface scattering (light goes thru and bounces under the skin a bit) and 2) some infra-red emission (the responsible for making some people appear magenta on digital photography)
So even when you get the palette right, to paint it like a believable material in a convincing way is still super hard!
So glad to see Nyma getting recognition! I went to elementary and middle school with her, happy to see her succeeding.
Could you explain the first step, "Manually Labeling Colors ? What did you manually label? I didn't quite follow.
From the way the article is written, it sounds like you presented yourself with a bunch of faces of completely random colours and then manually sorted them into two categories, "plausible skin colour" and "implausible skin colour".
But I feel like I must be misunderstanding, because I don't see why you would do that. Why wouldn't you gather photographs of actual faces and sample colours directly from the photos?
This part ( https://toneyalexander.github.io/inclusive-color-space/#pick... ) is especially cool.
Would love to see this 3-D picker ability in the many character designers out there, instead of the lame 8 swatches or whatever.
I think the ideal would be to use this picker to select a few tones, and then gradient map those across different parts of the body according to a grayscale map. Then you could fine-tune things like palm color, cheeks, facial tone, etc. individually.
You mention a bit about the latent social forces embedded within technological development and how that has related to skin tones and the technologies of color reproduction, so I figured I would share some references:
https://canopycanopycanopy.com/contents/color-goes-electric
The Kodak “Shirley cards” and test shots are of course famous in the media theory world for how they baked notions of race and “standard” or “normal” conditions into the medium of photography. Not dissimilar from the various discussions around facial recognition from approximately 10 yr ago.
I love this! Such a beautiful use of some real nerdy math to benefit humans in a way non-nerds would appreciate.
Very neat project. Could I use this code to generate skin tones in my own project? Or more meant to just be educational/inspirational? I guess I'm wondering if you've considered adding a license.
Good catch, added MIT license - feel free to use it! On top of that, my understanding is that math/equations are not subject to copyright, so the core part of this work (the equations/ideas) I think legally has 0 limitations - even if there were a restrictive license. But IANAL so might have some misconceptions.
Procedural generators are often free if you look around for a bit.
https://kuhantilope.gumroad.com/l/ultimate-skin
https://extensions.blender.org/add-ons/mpfb/
https://github.com/alessandro-zomparelli/tissue
ymmv, as some paid plugins actually speed up the funky blender workflow (the auto-Bake plugin by the same author is important for procedural generated textures.)
https://sanctus.gumroad.com/l/SLibrary?layout=profile
Have a great day =3
I'm using an iPhone Safari plugin called Noir to force darkmode and your tool is making white the absence of color. In other words, I think white is just an alpha channel.
Nice. This is relatively old work (2020) but Pinterest had/has skin tone features, for example if you're searching for cosmetic products. Worth a check if anything might be useful for your work. https://medium.com/pinterest-engineering/powering-inclusive-...
For anyone confused, turn off your dark mode extension. :)
Oh wow, didn't even think of how that sort of thing might affect the page.
On the movie Dolemite by Eddy Murphy, they go over the fact that black people absorb light and white people reflect, and how that changes the motion picture shooting process totally. Also how one size does not fit all. Cool project! I am not much into the humanities part of the project. For me this is as natural as having a localization framework, just in this case for skin tones.
Feature request: drop-down list of ethnicities, where the selection turns on specific guard rails in the parameter space (or retains full movement of the controls, but scales the values). Or is that too racist for 2026?
In the past we had to use multicultural crayons.
https://econocrafts.com/products/crayola-colors-of-the-world...
Related, pencils for more accurately drawing skin tones: https://hautfarben.org/en/products/skin-tones-pencils
I really like the idea of treating a color space as a task-specific optimization rather than something universal. It makes me wonder if we're underusing domain-specific color spaces in procedural generation in general.
Never seen anyone #E6FFF8 (bottom right corner).
An albino that consumed too much colloidal silver for years?
Setting R² to various large values like 50, 500 and 5000 then moving the slider around seems to give a decent and simple visual representation of what's happening.
This is excellent work! You’ve got a 49 year old and a 10 year old equally into it. That’s impressive - nice work.
Not my domain though I followed the data science enough to nod as if I really followed it,
but this is just lovely and inspirational work. A unique marriage of hacking in the celebratory sense, thoughtfulness, and meaningful contribution. Outstanding.
Wow what a touching compliment, thank you so much!!
Skin is so interesting as a surface to think about how light passes through and bounces off of.
There is at least one article published by a technical artist at Pixar who describes modeling a skull within a human scan so that subsurface lighting is more accurately rendered.
I suspect you could combine this with dr_scully’s comment and GolDDranks’ to reach the nuance involved in long tail realism details.
There's a lot of really good research papers that touch on this area from what I saw - would definitely be cool to see if that extra nuance could be added without dramatically increasing the math complexity; in its current state, the code could be used in any programming language without a library.
Cool work!
Me after I spend a day at the beach: #BA3A17
If you're ##CCFFFF please go to the hospital
Very interesting, I'll need to read it again lol
Seems to underrepresent pale skin tone variety
yeah, skin tone variations often feels it's about limiting lighter tone options while adding luminosity options for the exact same brown color. There seem to be B channel differences across darker colored people, and I wouldn't be sure if they are better represented either.
Turn off your dark mode reader extension
Under-represent relative to what, exactly?
Can't express redhead for example. Redhead being not the hair tone, but the translucency of skin that gives the look.
Translucency is not a property of colour itself.
This hits different after a long week