Sony A7 (m2), color calibration module, sometimes quite different colors compared to legacy whitebalance, not always

_DSC4665.ARW (47.3 MB)
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

(Hope I did the CC thing right).

I encounter something where it makes me go ‘huh?’. And I wonder if it’s a limitation of the module, me not understanding something, a bug, me not using it right, all of the above or whatever :).

The shot was a quick-n-dirty snapshot at home, and not even on of the keepers. But it illustrates the problem nicely.

If I use the ‘legacy color workflow’ (As in, I use whitebalance with no ‘color calibration’ ) I can get something that looks quite OK easily.

The whitebalance ‘as shot’ is not perfect, but if I use the white-balance module and the picker to sample the entire image (or the white part in the bottom-left) I get something acceptable:

The skin tone is quite ok, and the colors of the hat and the blanket match what I think of them in the real scene. I circled where I sampled, and I’ve got almost no other modules enabled except ‘the base starting point’. I disabled filmic and color calibration to just focus on the color part for now.


Now, when I set whitebalance to ‘camera reference’ like it was set initially, enable ‘color calibration’ again, and sample the same area, I get this:

The skin tone went a bit magenta. There are brightness changes , but I don’t even care about those (yet). It’s the tone of the skin.

And wherever I try to sample, or however I try to change settings and sliders in the CAT tab, I can’t seem to get it right to be more like the normal white balance mode.

Where is this difference coming from? Is it expected?
Can I do something to prevent it, or is this just one of those pictures where I need to discover to not use ‘color calibration’ ?

Have you tried the color temperature slider? It only seems to be a little bit off. Maybe you could try another Mode than CAT16,? Like Linear Bradford for example?

I don’t know but can see what you are getting. Perhaps it’s to do with the lighting, I’m guessing indoors with artificial light. Using XYZ as the adaption method improves things somewhat.

There seems to be an exposure difference, too. Did you change that, or is that a side-effect of switching the WB method?

Side effect of color calibration. All other module settings are untouched .

Yeah, ‘somewhat’ is what I noticed too but it’s still off.

I’ve never seen that happen. Can you double-check you have no RGB (channel mixer) values set by accident?

It’s quite odd to be honest . Warm skin tone vs strong magenta cast while the other colors aren’t that different.

Changing the temperature will change tolhe entire picture and still won’t get the skin tone in the ballpark of the pure white balance version.

I managed to get somewhat close by using a color picker on the skin on the white balance version, and then switching to custom mode in color calibration and massage the hue and chroma sliders for quite a while until I was close. And by then the white of the bed was quite off and it still didn’t look right .

And that was with using a reference to work towards , no way I can do the same with just ‘editing’ the image. The fact that a single part of the image yields quite another hue while the other parts are similar baffles me a bit .

I see it quite often to be honest.

‘Discard history → turn off filmic → turn off highlight reconstruction → turn off color calibration → white balance picker on the bottom-left corner’
‘Discard history → turn off filmic → turn off highlight reconstruction ->color calibration use picker on the bottom-left corner’

_DSC4665.ARW - colorcalib.xmp (6.7 KB)
_DSC4665.ARW - whitebalance.xmp (7.1 KB)

If I set the whitebalance module to ‘as shot’, and then use color calibration picker on the same bottom-left corner, the results are much more in line (you have to ignore the warnings though).
Seems as the ‘camera reference’ isn’t true, or is causing issues?

Could be. See

Hi @jorismak,

And I wonder if it’s a limitation of the module, me not understanding something, a bug, me not using it right, all of the above or whatever :).

Absolutely… :slight_smile:

To try to pinpoint the culprit, I played around with your .ARW:

(a) I converted it into a .DNG, which…

(b) I imported into daVinci Resolve:

(c) I set WB at about the same place as you did:

(d) And finally selected AutoColor:

Interesting comparison, don’t you agree?

Have fun!
Claes in Lund, Sweden

So basically I’m f-ed because of Sony weirdness … That would suck.

I don’t think I have a good setup to do the 'find your own multipliers ’ test in an accurate way.

But I could at least try and mess around with other settings in the white balance module and THEN try to sample color calibration to see if i get better results . And then remember the white balance setting :(.

Did other people find the correct multipliers for their Sony alpha cameras ? Can’t imagine there be much difference in the first few generations…

1 Like

seems DaVinci gets it even more wrong. The bed itself is still white, but the skin-tone seems weirder and the blanket is now really green instead of blueish / teal?

The ‘auto color’ seems to do an auto-levels / auto-contrast and nothing else.

Just a wild guess but any chance at some point you messed around in your camera with the WB doing a manual WB or something…could you have tweaked the color channels in camera?? I recall a similar thread quite some time ago and in the end the culprit was some setting in the camera that had been tweaked and forgotten about… anyway just thought I would ask

Since it’s RAW, any kind of those changes would have no effect on the raw data.

The only change i can make is a white balance ‘adjustment’ (push the auto WB more towards a certain color ) and that is not the case, the settings are all stock.

I have other photos (in a more normal lighting situation ) where color balance seems to work fine .

But in this 'low light at night’s lighting situation it seems to go wack.

Like I said in another post, if i change the white balance module to be around 2700k instead of 6502k, the color balance module seems to ‘jump to live’ and works nicely.

But those same multipliers on other daylight photos seem to have other problems , so it’s not a ‘one fix for all’ solution.

It almost seems like my camera has two ‘reference whitebalance’ points depending on the shooting situation ? Like in daylight 6502k is fine but in lowlight photos I need to use something closer to 2700k…

Interesting…ya I thought some cameras had adjustable cmos gains but maybe not commercial cameras, our digital nikon stuff in the lab can do that and I just had a faint memory of someone setting something in their camera that was affecting what was being read in for wb… i am likely confusing this with another issue someone had…

_DSC4665.ARW.xmp (7.7 KB)

Setting the input color profile to “linear Rec709” and setting the WB with color calibration to the white parts of the bed gives a more pleasing skin tone in my view.

Congratulations to the parents!

But this means your not using any profile / matrix for the camera , assuming the camera delivers something sRGB/709 like out of the box . That cant be right.

It does show that it’s something that’s weird in the profile Darktable uses for the Sony camera’s, as stated in the color balance docs.

If i look around online for another profile (i know they exist ) can I try to use that ? Never looked at profiles in use with Darktable now that I think about it.

Nice catch, @Thomas_Do!

@jorismak This is an interesting problem/question/dilemma :slight_smile:
It would be nice to find a valid explanation.

A crazy idea: could it be that you were recording a movie of the youngster, and at the same time shot some stills? Normally, an ILCE-7M2 would use picture profile PP2 (Gamma Still, Standard/Neutral) for single shots — but for movies they offer a lot of other settings, ITU709, S-log, &c &c &c.

Quote from Sony:

Since the parameters are shared for movie and still images, adjust the value when you change the shooting mode.

That is why I converted your .ARW into .DNG, to see if any clue would pop up (No, it didn’t).

Have fun!
Claes in Lund, Sweden

True. And I did not say that it was the right way, it was just an observation.