The last two numbers of the Kelvin WB tags are R and B multipliers, respectively. By convention, G is anchored to 1. I’m going to guess the first number is a temperature, not sure about the zero, as that’s not a valid multiplier.
Temperature is really just a function of the light at the scene, once you encode the image it becomes about the multipliers. Software attempts to give you a temp/tint control, but that has to be converted to RGB multipliers for infliction on the image.
Thank you, Glenn!
Do you think the WB on the images above is correct, or it’s a bug? (what is your opinion?)
Exiftool said 12 and didn’t have a name for the setting. I was wondering which it was.
You should test with natural or other base setting. Satobi may well do all sorts to tricks. It’s meant to produce warm desaturated sort of old film stock tones?
I’m not a “temp/tint” person, so I’ve avoided comment. IMHO white balance is a very scene-specific aspect of photographic capture, and some scenes vex a single setting. I tend to rely on my camera’s auto mode, and then I’ll tweak the multipliers if I want something different.
The image being discussed above seems to have plenty of neutral tones with which to set the white balance; I’d be relying on that more than some kelvin setting.
So there, you got the curmudgeon going… 
Okay, I’ll try with Natural and Flat (not sure what is better).
I have one more question. You recommended next workflow:
Input profile to: Use embedded
Auto Matched tone curve
Camera WB
Increase exposure a bit (Pentax matrix has exposure added)
And I would say “Use embedded” Input Profile with Auto Matched tone curve works pretty good. and with embedded profile WB - looks better. - Thank you!
However I would prefer to have that “embedded” profile as a profile-file, something like embedded.dcp (just in case, to be sure that I always can rebuild exectly the same jpeg).
I’ve tried to use exiftool to get profile and it looks like DNG file doesn’t have embedded profile (but RawTherapee use it). So I am a little bit confused. Do you have any idea, where RawTherapee take it, or how it builds it? and is it possible to get that embedded profile, I don’t know, build with dcamprof or anyhow else?
I downloaded the DNG, found it has the following tags:
[EXIF] CalibrationIlluminant2: D65
[EXIF] ColorMatrix2: 0.8552093506 -0.2741088867 -0.1199645996 -0.3611755371 1.220397949 0.1549835205 -0.09745788574 0.1833190918 0.683380127
The D65 ColorMatrix value from a DCP/DNG for a camera model is what RawTherapee usually uses for the internally-constructed camera profile; from RT camconst.json I got this:
{ // Quality C, possibly the same as K-1
"make_model": [ "RICOH PENTAX K-1 Mark II", "PENTAX K-1 Mark II" ],
"dcraw_matrix": [ 8596, -2981, -639, -4202, 12045, 2431, -685, 1424, 6122 ] // DNG
},
The RT numbers are encoded as dcraw coefficients, which the DNG coefficients can be compared by multiplying each number by 10,000. Note the camconst matrix has the comment ‘//DNG’ appended, and those to my knowledge usually come from one of the Adobe DCPs in their DNGConverter application. I went to look that up, and found this from their Pentax K-1 Mark II Adobe Standard.dcp using exiftool:
Color Matrix 2 : 0.8596 -0.2981 -0.0639 -0.4202 1.2046 0.2431 -0.0685 0.1424 0.6122
Ta-Da.
So, if you want to use the “embedded” profile from the camera’s DNG metadata, you could paste the ColorMatrix2 numbers from the DNG into RT’s camconst.json file over the Adobe numbers. Actually, there’s a way to make a local .json file that replaces such without munging camconst.json; look that up and try it.
I did just that in rawproc, my hack raw processor, that lets one paste a dcraw coefficient collection in the camera profile box. Here’s a snapshot comparison:
You’re looking at a rawproc program screenshot, with a rawproc snapshot embedded so it can be compared to the rawproc display. The rawproc display is using the numbers copied from the DNG, the snapshot uses the Adobe/RT numbers.
I don’t see much of a difference in white balance in your first screenshot @Shutovds. Maybe it’s because my monitor isn’t calibrated, so I used the vectorscope for an objective comparison. First one is of the left image (the JPEG) and the second one is of the right image (RawTherapee)
There actually aren’t many neutral colors in the image. The parts you said are neutral have a slight golden tint in both the JPEG and RawTherapee output. The asphalt is more neutral than the sidewalk. I put my cursor over the asphalt to show its color in the vectorscope. Notice where the small circle is in both vectorscopes. They both show a slight yellow tint. I think what you are seeing is just a difference in saturation.
Thank you Glenn!
I found the camconst.json file in /usr/share/rawtherapee/ is it correct that I need to copy it to /home/my-home-dir/ - “next to file options”
and after that edit the line:
What is “ranges” and “white” for Pentax K-1? does it affect WB? Should I edit it anyhow, or “dcraw_matrix” should be enough? (I saw a lot of instructions in the file, but not sure that without experience I can set up all possible settings properly, and also not sure what should be set up)
I may have missed a nuance to the discussion here and I know its about WB but WB is being suggested as the reason for the image appearance and this still could be but I find that RT has color preservation models …I think the film one is the default used with tone curve/automatched tone curve … and it give one look but for me when I extracted the preview from the provided DNG… opened it in xnview and then opened the DNG I could see the strong difference however for me simply changing the profile to embedded and using either luminance or saturation and value blending made the images much much closer… my point is there isn’t just one look to the image preview from the color profile and wb…those settings in the curve that is often applied can dramatically impact what the resulting image looks like out of the gate…
Default processing started here with pretty vibrant green
Then change to embedded (small but I think in the direction of the tif preview)
Then changing away from the film color model to saturation and value blending
Or luminance…
Were closer to the preview …not other edits including exposure were applied in any case.
Thank you, Lawrence37! I think you are right, because when I followed @nosle advise and choose “Use embedded” input color profile and Auto Matched tone curve it becomes close to in-camera jpeg. So I agree with you. I also updated “dcraw_matrix” in camconst.json so now it is much better.
I still have a question about Auto Matched tone curve. When I choose it, by default it switch the tone curve to Flexible Film-like curve. Is it the best option for color accuracy (I mean Film-like curve) or it is better to switch to perceptual? (I am not sure why, probably I read on this forum, but I think that Embedded color profile “better works with perceptual”) So what is the “recommended” approach?
I don’t think you need to copy over the whole camconst.json, make one with just the section you need on the K-1 Mk II
The “ranges” section you call out is about different white levels for mapping the camera measurements into the display/export integer ranges, nothing to do with white balance per se.
Based on what I’ve seen, I’d expend the effort to make the local camconst.json with the DNG-extracted dcraw_matrix for your camera. Adobe is known for taking license with the values, and I personally prefer the DNG-extracted ones based on my predilection for more neutral colors. But, I ended up making my own camera profiles from spectral measurements, so I’ve long left the matrices behind…
Thank you, Glenn!
About the own camera profiles - I’ve tried to use dcamprof with x-right color checker, but I don’t have the devices for spectral measurements and for setting up the lights, and without that I am not sure that the profile I’ve created is accurate. I tried to follow the instructions, but actually the light is not D65, D50 or A, it is something in the middle of I don’t know what…
Can you recommend, how to build the profile more accurate? What devices might be used for that?
Are you running windows or linux…you can do a lot of custom dcp profile editing in the adobe profile editor… dcamprof is good but its a CLI and you need to sort out all the myriad of options, and you need an expensive version, ie lumariver to get this power in a GUI… As an alternative this is pretty powerful…you can do spot color tweaks, wb, primaries,tone curve and use the color checker…
You can save the the embedded profile from your image and you can also check the other adobe ones for eg…
Note the profiles that you can check…
Its free and needs no connection to Adobe…
Its easy to use…
dcamprof will generate a whole slew of reports with the -r switch. In the reports subdirectory, the patch-errors.txt file will give you dE values that characterize the difference between each patch reference and what your profile comes up with for that patch. For matrix profiles, dE < 5 is good.
(Folks will be tired of hearing this from me…) I spent about 2 years working on a way to collect spectral data (SSF) on my cameras in order to make LUT-based profiles with dcamprof. The gold-standard for such is to use a monochromator in an optical lab environment with other instruments, many thousands of dollars US; I came up with a cheap alternative using a diffraction grating in a wooden box, about $130 US, and my profiles were within a half a max dE of the lab product.
I looked through my collection of SSF data for your camera, closest is data for the K-5. Nikon and Canon tend to make their Bayer filters consistent across their recent camera lines, don’t know such about Pentax. I used a Nikon D7000 SSF profile with my Z 6 once just for grins, worked okay. There are a number of reasons SSF-based profiles are better/more convenient, discussion for a separate thread…
I think at this point I’d spend more time getting a decent matrix profile for your camera than dorking around with SSF profiles, that is, until you take a shot of some extreme colors and wonder why the hues are all posterized. That’s what got me going…
Seems the problem with WB has been resolved! 
After I changed the “dcraw_matrix” in camconst.json file:
RT show relevant values of WB:
It is possible to see, that chosen White Balance method is Camera and Temperature: 5487K, Tint:1.015. That parameters much more relevant that were before the “dcraw_matrix” was changed (Temperature was 6467K, Tint was 1.014). The image looks neutral.
Thank you, everybody, thank you, Glenn!!! @ggbutcher
Probably “dcraw_matrix” should be fixed not only in my local settings, but also in RawTherapee software, in order to all Pentax K1 mark II users have good White Balance settings.
For DNGs, “Use embedded” means using the color matrices in the DNG. The color rendition will be the same as if you add the matrix to camconst.json.
I think luminance and perceptual are all fairly good at maintaining accurate colors. Film-like tends to increase saturation as it adds contrast. There’s no “rule” about which curve modes work better with which color profile. What you hear from other people is based on opinion and personal experience. If you prefer a more neutral look, luminance and perceptual are the best options. If you want more vibrant colors, film-like is probably better.
It’s annoying that you can’t* have auto matched and any option other than film-like.
I have set up dynamic profiles to achieve this but it’s unreliable as the auto matching seems to happen when the image is opened so you can’t batch.
I will not tired
I have no idea about how to collect spectral data, and how to use diffraction grating for that. Can you share your experience, or/and recommend anything to read about that?
(my level at the moment is: lenear and several curve profiles with dcamprof with xrite color checker 24. As I understand dcamprof allow to build targets with bigger number of colors, but to use it I need some kind of device to measure the colors of the target and I don’t know what device to use for that)
And you’ve intrigued me about your collection of SSF data. Why the data for the K5 is closest?
There are a number of reasons SSF-based profiles are better/more convenient, discussion for a separate thread…
Yes, seems the problem with WB was resolved, do you think it is a good idea if I start the separate thread for discussion about color profiles?
@Lawrence37 I agree with @nosle it would be great to have a choice of options for Auto-Matched Tone Curve in dynamic profiles
