Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: A Color Palette Generator I Built For My Senior Thesis (colormoo.com)
109 points by joshrael on Apr 9, 2014 | hide | past | web | favorite | 44 comments

Picking colors is an interesting problem at the boundary of what can be done algorithmically with satisfactory results.

For my latest side project, I needed sets of up to 5 colors that go well together but are sufficiently distant not too be confused. Since this is the task that palette generators typically set out to achieve, I tried a couple of them [1,2], as well as some "hand crafted" palettes found on design blogs [2,3]. In each case, I found that the palettes, while satisfying the desired properties, were aesthetically not particularly pleasing, in a sense lacking a unifying quality.

Thinking that "picking colors that go well together" must be a solved problem, I resorted to looking up photographs of 20th century paintings (e.g. [5]) and using the color picker in Gimp to extract what I visually perceived as the most important colors. This gave me far better palettes that any of the generators.

There are ways to do this algorithmically, as a form of constrained color quantization, and I later found online implementations as well [6] that give results very close to my hand picking.

[1] http://colorschemedesigner.com/

[2] https://www.colorschemer.com/online.html

[3] http://flatuicolors.com/

[4] http://niklausgerber.com/blog/flat-ui-color-autumn-edition/

[5] http://i1.ytimg.com/vi/XcwZF1aF4LM/maxresdefault.jpg

[6] http://labs.tineye.com/color/

> at the boundary of what can be done algorithmically with satisfactory results

Mostly due to insufficient models of human perception of color. Our perception of color is not accurate, so picking complementary colors based on their actual wavelength will be less pleasing than picking complementary colors based on a perceptual model. I don't think any of the pickers you link to attempt this.

One attempt with libraries in a few common languages is HUSL: http://www.boronine.com/husl/syntax/#997061

A much older approach is described here: http://en.wikipedia.org/wiki/CIELUV

Learn more about color perception here: http://en.wikipedia.org/wiki/Color_perception

[Edit: just noticed this was referenced by this comment below: https://news.ycombinator.com/item?id=7562905]

Take a look at Munsell for Designers[0] - it purports to solve the UI palette problem.


I quite like colourlovers for palette ideas and construction - they have a tool that extracts palettes from images http://www.colourlovers.com/photocopa that might have helped you.

Not connected to the site except as a user.

Here's a shameless plug for http://www.colr.org . No palette generation algorithms, but lots of tagging and classification.

I wonder if you could train a machine learning algorithm on 20th-century paintings? ;)

Does the algorithm use HSL as its color space? I've seen many online palette tools that use HSL color wheels for picking colors. It never made sense to me why they wouldn't use a perceptually uniform color space like CIELUV. I actually made a color space derived from CIELUV specifically for the task of picking colors [1]. I never wrote a 'proper' palette generator using it, but I did make a demo that works pretty well [2]. Considering that you're actually studying this, I would be curious to know what you think.

[1] http://www.boronine.com/husl/

[2] http://www.boronine.com/husl/syntax/

Thank you for creating HUSL! I discovered it as I was midway through my project. I am very glad that this exists! I agree, CIELUV or CIELAB would probably be a better fit. As I mention in my comment above, it wasn't immediately clear to me how I would generate a palette using the space given the holes, especially since I allow the user to choose any color they want.

I am curious, how would you approach generating a palette using these color spaces?

Thanks :)

> how would you approach generating a palette using these color spaces?

This is exactly the problem I was trying to solve in creating HUSL.

You can think of HUSL as a version of CIELUV (Or, more specifically CIELCHuv, the cylindrical transformation of CIELUV) that is stretched vertically to fill in the 'holes'. HUSLp, the pastel version, is similar, but instead of stretching, it cuts the saturated colors off. (You can switch between them in the demo).

In fact, those stretching and cutting functions are the only thing that HUSL really does, the rest is all CIE math :P

I just checked out HUSL, that's amazing. The syntax demo is really cool. The colour schemes it generates are much more visible and pleasant to the eye (my eye, at least) than Solarized.

Disclaimer: I'm slightly colour-blind - take with a pinch of salt :-)

HUSL is awesome. Wrote a library last year that uses it: https://github.com/nluqo/autumn

I like it but isn't blue under represented?

It could be. The only new component that HUSL defines is saturation (it's a distorted version of CIELCHuv's chroma). Both hue and lightness are exactly like CIE defined them after doing actual empirical research on color vision.

So while the hue component isn't perfectly perceptually uniform, there is no way I could improve it without doing an actual study.

I generated about 30 different palettes just to try it out, but I didn't like any of them. Most of the palettes generated had colors that didn't even work well with each other, maybe it's just my taste. Good work though, I'm sure you'll be able to make it better.

Maybe this is useful, but I'd like to see a little bit of info on each palette as to WHY it was generated the way it was generated. The palette you need for a pie chart is much different than one you need for a website.

For web site palettes, I like to poke around on http://www.colourlovers.com for inspiration. http://colorschemedesigner.com helps with finding appropriately distanced colors that work nicely together (set colorspace to Opaltone colors to find a nice palette) and there is always http://www.colorhexa.com when I just want to quickly darken or lighten a color.

Sometimes I'll browse through designs on http://dribbble.com/, when I see a design where I like the colors I'll click on it and it'll actually show me a nice palette of all colors in the design. What's nice about it is since these designs were done by great designers the palette almost always works as-is which is quite nice. :)

Here is my immediate suggestion: Show only one palette at a time. By putting incompatible patterns so close, you get a matrix of colors that with high probability do not harmonize well.

I am assuming you have a slightly different algorithm for these three, since the middle one doesn't seem to match the root color's brightness, and they seem to match saturation differently too. Make a choice between these before you generate, and if you want to eliminate this choice, I would go with the one that tries to match brightness/saturation as much as possible.

This caught my eye because I developed my own algorithm for this a few years ago and applied it to syntax highlighting[1].

[1] http://sweyla.com/themes/

Hey, I saw your page a while ago! Actually, I ripped it off to demo a project of mine [1]

[1] http://www.boronine.com/husl/syntax/

FWIW, Web of Trust was blocking your site due to poor Trustworthiness and Child Safety Warnings. See here: https://www.dropbox.com/s/2ol9ad09qv13jk9/Screenshot%202014-...

That's pretty rad. Is the source available anywhere? I'd like to see something like this for Sublime.

You can download the TextMate theme file for the colour scheme you generate. You can use the tm file directly in Sublime.

Oh nice, I wasn't aware of that. Thanks!

Thanks for the feedback everyone. Perhaps I should have included a bit more explanation.

I was primarily interested in an algorithmic approach to generating color palettes. I am familiar with the various tools available (Adobe Kuler being my favorite) that use color theory as a basis of providing colors to go with one that you've selected (i.e. complimentary colors, analogous colors, triadic colors). One the palettes that my app provides uses this approach.

Another approach I looked at was a trick that I've seen many designers use in which you "mix" a little of the color you want to use with another color. To play with this concept, I generate a random set of colors and blend it with the one that you've selected.

Finally, I am a huge fan of Gregor Aisch's work [1] with colors in regard to statistics. My third palette borrows a method of his that I had not seen elsewhere. He essentially maps a particular color space into two dimensions and keeps the third constant. You can sample equidistant colors along a line in that space to derive a set of colors that look good together. In my case I used HSL, where I hold Lightness constant (I use the value of lightness of the color you selected).

As he mentions in the post I've linked below, HSL is not necessarily the best color model to use this way. A more perceptually uniform color space like CIE LAB would be better. However, working with this space in this way is a little more complicated because, since this color space models how we perceive colors, it has holes. I had trouble finding an elegant way around those holes such that I could consistently offer a fourth palette.

Anyway, after looking at each of these methods, I was curious if any one of them would rise to the top as the favorite among users. Or, perhaps one method would perform better for some colors over others. I am in the midst of processing all this data and my thesis should be published within the next month. Thank you for helping me collect data!

[1] http://vis4.net/blog/posts/avoid-equidistant-hsv-colors/

Very cool, however I was hoping the cow's spots would change colors as I moved the sliders.

Kuler [1] is a great tool by Adobe that I love to use

[1] https://kuler.adobe.com

Is this the same as Munsell for Designers[0]?


Can this create a palette of an arbitrary number of colors, or is it limited to six? I've been looking for a good way to generate an arbitrary number of visually distinct colors for a while, to use in visualizing high dimensional datasets in bioinformatics. The closest I've come so far is http://tools.medialab.sciences-po.fr/iwanthue/ (which is great but I need to port it to R one of these days)

I have recently written a library that picks N as distinct possible from the RGB color space. Since RGB is not a perceptual color space, it measures distances between RGB colors in CIE Lab. Since the optimization of the most distinct color set can get stuck in local maxima, it uses simulated annealing for optimization.

One nice touch is that you can plug your own random RGB color generators that feed the optimizer, to find N distinct colors in a subspace of RGB. E.g., I use it to generate distinct pastel colors in visualization of linguistic data.


(I am not an expert in this topic, but I couldn't find a good Java library with solid theoretical underpinnings.)

For arbitrarily many colors, use the golden ratio:


wow, this is great! I've been trying to find this for a while.

Their theory page (http://en.wikipedia.org/wiki/Lab_color_space) goes into how they do it, it's a pretty interesting approach of pre-calculating using CIE lab and then hcl for display/ui.

And the ui is well done, too.

For those that want to do this on iOS, I created a handy UIColor (or NSColor if OSX dev is your thing) category that can create 5 tone color schemes by passing in one color.


Beyond the color scheme part, another cool thing you can also do is find distances between colors using CIE_LAB specification. This is great for finding "like" colors on the fly.

Not working in Chrome on OS X. Seems to look ok in FF and Safari although I am not sure where the input for color is supposed to be positioned, it looks out of place.

I see you used Zerb Foundation, how did you like it I am thinking of using it on a side project.

Edit: Working in Chrome after a few refreshes.

Thanks for noting that!

I'm a big fan of Foundation, although if I wasn't trying to get this together so quickly I might have used something like Skeleton[1]. But I'm sure that is another discussion thread entirely ;)

[1] http://www.getskeleton.com/

Cool tool. Good work! Feedback: I agree w/ other comment...the palette options I get don't seem to go well together. Maybe you're trying to include too many colors in a palette? Also, I'd center the checkboxes on the button when picking a palette.

If you're into this kind of thing, definitely check out the NASA color usage research lab website: http://colorusage.arc.nasa.gov

with all due respect, and only for whatever it's worth, every palette it generated is worse than any palette i found on colourlovers.com. but cool project. :)

Proof, I suspect, that humans are still better than computers. Or at least better than my computers. For now.

I posted something similar a while back...


Your CSS is a little broken in chrome v34, the black box that overlays when you mouse over colors are about 15px to low and kinda screw things up the lower down the page you go...

Great job, I like it a lot. Congratulations on graduating soon. I wish you luck in all your future endeavors :)

I'm getting a lot of really ugly palettes... by "a lot" I mean all of them.

This may be a presentation problem as generally when using a palette, colors are not used all equally and not necessarily all right next to eachother. Perhaps a thesis for another grad student.

Nice work! I made http://cy.mk/ as a project for learning javascript a while ago. I need to go back to it and work on it some more.

Neat; glad to see HSL values included!

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact