This depends on the purities of the complementaries. If you take the sRGB blue and yellow (for the sake of an example) there is a vast difference in the “perceived brightnesses” as well the luminances (I think MacAdam writes “brightness” when meaning “luminance” because the unit is millilamberts).
Now, you would be able to just add some magnitude of the neutral stimulus (1, 1, 1) to the patch with lower luminance (blue in this case) to bring it to the same level as the complement. Those will still add up to neutral, but the blue will have reduced in purity.
Thank you for the motivation. In this case with the low luminance of blue there does indeed not exist a complementary sRGB yellow which is dark enough. So the addition of white to the blue does the job. Here are the modified values:
(051,128,204) original blue.
(131,171,231) is the brightened blue instead of (051,128,204).
(199,167,000) is the unmodified complementary yellow.
From the above numbers it is clear, that the yellow (3) cannot be darkened because of the 000 value. But the original blue (1) still has some room to increase (2). With this, the luminance of (2) and (3) are identical and have the value L = 39.7 %. The luminance of (1) is 20.5 %.
Please correct me, if I am wrong. Now I wonder, which website and tool does this luminance-balanced color harmony correct. And a pitty, that quite a lot of sRGB colors have no “visible” counterpart.
And here is finally the CIE-1931 chart with these three example colors. You can see, that only colors inside the same circle can be luminance-complementary.
Please note:
The black axes. grid and circles are cropped screenshots from figure 1 of the original MacAdam paper.
The small color circles are color (1, 2, 3) from left to right. The opposite colors are in harmony because the connection line goes through white and in this case through gray 39.7 % = (169) = #a9. And the colors on the same circle are in luminance-harmony.
The sRGB triangle in the background is my own Processing calculation with absolutely correct sRGB colors in the xy-area. Most colors of sRGB triangles in popular CIE-charts are wrong.
The three circles are darker than the sRGB triangle in the background. This, because the triangle is based on 8 bit colors with one of the three RGB value always at 255. In total 3 x 256 x 256 = 196698 possible triangle color points in this plane.
The color triangle has some colored straight lines. Guess what they mean !
But finally the brain and the emotions decide about harmonies and not (only) math or models. May be in other countries or on other planets the inhabitants feel totally different about harmony?
Much too late I understood, that Munsell is one sound and serious approach for color harmonies and not RYB or Itten or a lot of others. May be the RYB vectorscope and my school paint education placed my on the wrong path.
From the previous MacAdam circles you see, that it is directly related to the irregular shape of Munsell’s color system 3D model. Not only in sRGB color space. On Munsell color system - Wikipedia you see this color harmony table:
If the chroma turns higher, more and more colors do not have a luminance-harmony complementary. Which depends in addition on the relevant color space.
My random chosen example of blue/orange color is quite near to Munsell 5PB-6/6 and 5Y-6/6. I just had luck. My choosen colors are very near to 5PB-7/8 [141,174,239] respectively luminance-leveled 5PB-5/12 [46,123,212] and 5Y-6/14 [195,138,0].
Source: (415 kB) ConversionsBetweenMunsellAndsRGBsystems.pdf
For sure the implemenation of Munsell in the DT vectorscope is questionable in terms of CPU requirements, speed and benefit. Here is a screenshot with the 5 connected pairs of above Munsell named colors.
Is just the white point shifted in above RYB vectorscope for these Munsell colors? And is this the case for other Munsell complementary colors as well? And is such point allways at the same location? I cannot believe. At least challenging!
May be I should now better rename the topic of this post from RYB to Munsell?
It took quite some time for me to learn and understand more about color wheels, vector scopes and complementary colors. A big help with this are various very good Youtube videos of David Briggs. He does explain in the 30 minutes video on https://youtu.be/ZLIIYj2X-Qc about “Dimensions of Colour Today” and at at 21:26 in detail about various color wheels.
The RYB color wheel is indeed wrong and outdated. For Darktable v4.4.2 this does mean, that only Luv vectorscope does show correct complementary lines of RC, MG and BY through the white middle point. Vectorscope JzAzBz is also quite ok, but RYB is not at all correct. Please find attached an image with screenshots from Darktable and DaVinci-Resolve vectorscopes of a simple RGB test image.
For complementary colors (which have to add to white or gray!) the CIELUV vectorscope of Darktable is correct and of course also DaVinci vectorscope. In the CIE 1931 (and also the CIE 1976) diagram of DaVinci the correct complementary straight lines are also visible quite well.
As mentioned in my previous postings for “real” harmony, not only the straight line through white is important, but also the same perceived brightness. This second condition is not visible in the vectorscope and can not always be realized. See the circles in the CIE-1931 diagram of my posting dated Sep 27 with the MacAdam circles.
And here are the previously mentioned Munsell colors again with CIELUV vectorscope.
No need to draw dotted connection lines through white, because all five named Munsell example color pairs 5R-5BG, 5YR-5B … up to 5G-5RP are perfectly opposite aligned! And the other non-named 15 color pairs as well.
To summarize: No need to implement something like Munsell in Darktable. It is already there! You just have to use CIELUV vectorscope and never (EVER!) RYB.
@darky, aside from looking from a tech perspective, have you visually compared the same image edited using RYB harmony to a version using CIELUV RGB harmony, to see which you find more aesthetically pleasing?
I believe a dedicated “skin tone” line is unlikely to be added. The documentation for the vectorscope indicates:
The vectorscope does not have a “skin tone line”, which is a flawed generalization rather than a universal standard
So it’s not a lack of a feature, but rather an intentional design choice.
The current color harmony UI is able to place a line at any arbitrary hue angle, which should suffice for this use case. If you use, “monochromatic”, shift-scroll changes the size of the wedge, and ctrl-scroll changes the position by individual degrees. From what I’ve read online, most tools use a hue angle of 32-34°, but like the documentation states, this is a generalization, and not a standard.
The lack of a spec from the ICC or anyone else doesn’t make it any less useful as a tool. As arbitrary or “flawed” as it may be, if you push all the skin tones onto the line in Resolve they look good, end of story.
I didn’t realise that one of the views allowed you overlay lines though, that will work, thank you.
There is a skin tone colour measurement system called Individual Typology Angle (ITA°), expressed as a function in the CIELAB colour space, that’s used by dermatologists. Perhaps that’s a rigorous enough standard that could also be useful for guiding skin tone editing in our photos?
Google has done a ton of work as of late working with something call the Monk scale… Its interesting you can actually download their training dataset that takes multiple images of a person in different light and while smiling laughing etc to capture as much nuance as they can… I should pull the swatch in to DT and see what hue angles you get… https://skintone.google/
from left to right I get 30, 30, 40, 40, 36, 32.4, 23.8, 17.7, 26.3, 26.7 which has a mean of 30.29
out of curiosity, the patches in that ITA graph are 32.4, 30, 26.7, 25.4, 22.4, 17.8, 16.8 for a mean of 24.5
it does demonstrate that the “skin tone line” is a pretty crude approximation, but a skin tone curve or even a skin tone field could still be quite a useful tool in it’s place. It at least seems we have the information necessary to implement such a thing.
Ya a further read on the google stuff notes that their scale is more for generating perceptual awareness of skin tones so that differences are noticed rather than for matching skin tones… They also have a research paper link that I have not read yet…
The art of color, the color perception of the eye + brain and the science of color physics meet when it comes to color harmonies. This is a apparently a complex and exciting challenge!
The ColorAide python approach to colors recently updated on 1st September 2023 the following quite interesting web page: Color Harmonies - ColorAide Documentation On this RYB color wheel is mentioned. And the perceptually more uniform OkLCh color space is used as default.
Thank you Soupy for the idea to compare images “harmonized” with RYB versus CIELUV vectorscope. I have not done it yet. May be the result depends strongly on the individual image color distribution? And it certainly depends on the subjective experiences of artists and viewers with their personal color harmony feelings.
Thank you Nis Donatzsky for creating the feature request on GitHub.
Start a new thread, post a few images side by side, one with ryb harmony, one with cieluv, and without telling the audience which is which, have them vote on their preferred harmony. Would make an interesting experiment to see how subjective these things are, or whether the audience reached a consensus.
The old master painters reached a consensus on ryb which never would have occurred had it looked bad, though you might find greater variation in modern taste, following the introduction of many new styles.
Bruce William published last week a contribution on https://youtu.be/HyqteZsLZyI about color harmony with darktable. He admits to be no expert in color theory. I think it is a high risk and bad idea to present undefined RYB colors or sRGB-based RYB colors to unexperienced users. They take it for the only color harmony truth.
RYB certainly has valid and historical reasons with paint, but has it to be transfered as the first and only option to photographers in Darktable? I am still not clear, which sRGB colors are used in the RYB vectorscope of Darktable. Some unknown Itten colors? Currently sRGB Yellow (255 255 000) and Magenta (255 000 255) are exactly opposit (RYB harmony), but Itten did certainly NOT use these sRGB colors. And other sRGB colors like Red (255 000 000) and Green (000 255 000) are not opposite. The RYB vectorscope does look completely arbitrary for me.
The idea of Soupy about side by side images with “RYB” and “CIELUV” harmony for comparison is good, but I am sure, that any voting result will be completely dependant on the subjective selection of the image and its original color scenery. Is there any kind of reference photo which is suited to make a reliable decision about color harmonies? I fear this does not exist.
The photo of Bruce in his video with the skin tone and his blue or “harmony-green” shirt is probably not suitable, isn’t it? When switching from RYB to CIELUV or JzAzBz vectorscope the shirt will turn to cyan or dark cyan complementary (real opponent). I personally can not tell which shade of green or cyan is more pleasant to the eye. May be just keep the blue original shirt?
Or even worse, make the skin orange with RYB or yellow-green with CIELUV? Now I have to stop!
Anyway, colour harmony should be based on light, not pigment, so RGB should be used instead. The idea being, the complementary to red is cyan, so if you want to add a colour to 'balance the red", cyan will add the other colours of the spectrum, leaving your eye not wanting to see another colour as the whole spectrum is present. These are called optical complements.
!
Now I have to stop!