How to process mixed conditions lightning & getting nice colors in Darktable

I usually have a conservative approach to processing; there are a few instances however where I fear this doesn’t help – like in this example taken at evening in a cafe’.

I should point out that this is a snapshot that has only a personal value otherwise I’d have discarded right away. Anyway, it is a good example of my inability to “make” good colors out of an image shot at high ISO (camera is a Fuji XT2), preserving a natural rendition of skin tones.

I wrote “make” and not another word because honestly I can’t remember what was the real color conditions of that environment, so I don’t care about being honest to that particular situation – it’s not photojournalism! – but I do want to recreate nice, natural colors.

Please find here the jpg as I have processed it and the original raw file plus the xmp file from DT (oh yes, I forgot to say that I do my work in Darktable 2.7 so for the purpose of learning something new I’d ask suggestions on how to better process this image in DT).

20180922_FUJ8679.raf (25.4 MB)
20180922_FUJ8679.raf.xmp (5.3 KB)

License: CC-BY NC (https://creativecommons.org/licenses/by-nc/4.0/ )

Hi @aadm!

That’s what I could do with Darktable 2.6:

My xmp file doesn’t make sense for you since you’re using newer darktable version.
So, I describe what I did:

  1. white balance (window as selection)
  2. noise removed (denoise profiled)
  3. LAB curve (b channel - yellow in bright areas of the image moved down):

B%20channel

  1. Repeated the process with the curve, this time with the parametric mask:

  1. Increased saturation with Velvia

  2. Brightened up image with another curve

2 Likes

Boris! this is brilliant! I have never used the b curve shaped that way! I was in fact struggling to use the same few modules I normally use (tone curves in lab mode) to fix the colours but couldnt get anything “right” out of it. I will try to replicate your work, I wonder what is the effect of WB to “fix” the colours in this case…

Simple white balance correction. Light in the window is not as pink as in your example. I suppose it was daylight.:

WB

darktable 2.7 dev
20180922_FUJ8679.raf.xmp (24.8 KB)


A possible crop:

Mainly I’ve shifted the greens to oranges with the hue and saturation equalizer
20180922_FUJ8679.raf.xmp (181.4 KB)


darktable (2.6.2): 20180922_FUJ8679.raf.xmp (3.4 KB)

1 Like

My take:

20180922_FUJ8679.raf.xmp (5.0 KB)

I took the white balance from the chair, they look like being white :smiley:
And darkened a little the right corner to balance the photo and put the focos on where the action is.
The noise reduction is a preset I found somewhere in internet looking for darktable fuji x-t2 denoise (I think it was the dpreview forum). It seems to work quite well with the Fuji files…

@s7habo hey Boris you know I haven’t been able to replicate your results! First of all the white balance (but I don’t think it’s the key aspect here) – I used spot metering and draw a rectangle on the white area around the window, did you perhaps included the whole window with also a bit of frame (the wall around the window) like the screenshot you posted?

[this is my white balance settings plus spot area]

Screenshot%20from%202019-08-29%2000-06-25

Screenshot%20from%202019-08-29%2000-06-31

Also, did you increase a bit the exposure before applying the first curve? (I did, about +1ev).

Then step 3, when you say “repeated the process with the curve”, you mean you added a second tone curve instance, with same shape drawn on the b channel this time applying the mask? This is what I did but it only decreases the overall saturation quite a bit – plus my mask does not look at all like yours.

[this is my first curve applied, no masking]

Screenshot%20from%202019-08-29%2000-04-12

[this is the second curve, tryng to replicate point n3 of your list as I understand it, with resulting mask which is very sensitive about the input values to the left – increasing them just a tiny bit makes the entire mask go away because as you see the values are all below 0]

And Velvia? I bumped it up to 75% but I get no punchy natural colors like yours, my image is still flat!

Sorry but your version is the one I liked the most and my inability to replicate it is a bit disappointing!

I should thank everybody that has tried to improve the image, you all did much better than me; I especially liked @s7habo and @msd for the colors and mood of the image and @Daniel_Catalina for the Fuji noise reduction settings.

OK, let’s do it step by step. What’s important is that you understand what’s happening.

  1. White balance.

White balance is important to eliminate color distortions produced by the camera itself. In this case this is the magenta (pink) which you can see well in windows. Although this is not so prominent in the other areas of the image, this color shift also affects the other colors, including the colors of the skin. So, let’s correct that first:

Before:

After:

As you can see, I have selected a range in the window for whom I thought it should be white. Now you can see that the chairs and T-shirt of the girl look more natural.

  1. yellow color cast

Now it is a time to remove the yellow tint caused by the monochrome illumination. What is noticeable is that the brighter areas of the image are more affected by this than the darker ones.
So, if we want to remove the color cast, we have to take that into account.

The B channel of the LAB color curve is very suitable for this. First we have to find out which area we shouldn’t influence too much. This is, for example, the area of the woman’s face. Therefore we take color picker (area) and look where this area is in the B channel.

This will be our " break point " for the curve:

Now we remove the yellow cast by making a bend on the marked place in the curve, and make sure that the wall does not become too bluish.

This is a bit tricky here but the reference next to the face is also the light area of the wall to the left of the window. It should remain white as far as possible:

  1. Increasing color saturation with Velvia

Since the yellow tint has suppressed the other colors, the color saturation must be increased to see what else needs to be done:

  1. further color correction

As you can see, the facial colors are slowly appearing, but the yellow cast is not completely gone yet. We now repeat the process with the curve, this time with parametric mask to protect the face colors even more:

  1. brightening the image

At the end, we take another color curve to brighten the picture:

The whole thing is of course very tricky, and each time you get a little different outcome, but still, with a bit of patience you can get very satisfying results.

10 Likes

thanks for the very detailed step-by-step Boris, much appreciated all the time you’re spending to help me!

1 Like

DT 2.7 20180922_FUJ8679.raf.xmp (19.2 KB)

I’m not sure if the yellow color cast should be removed or whether it is part of the picture. Hard to say when you don’t know what the situation really was.

3 Likes

Couldn’t correct the colour so I decided to do my own thing.

RawTherapee capture sharpening and highlight recovery
gmic norm, local contrast, weighted smooth, brightness-contrast, resize

nice one, I would also in these cases go straight to B/W, but I kinda wanted to see how one would go and correct the colors --> that’s why my original post.

@aadm in case you want to play more with noise reduction, I saved at some point this link: https://tomassobekphotography.co.nz/articles/reducing-noise-in-Darktable.php

It is updated from time to time and you get a lot of options :slight_smile:

Ciao @aadm,

Here is my attempt. I did not concentrate on colour casts, but
on the crop instead…

/Claes

I don’t think so but thanks. It was a quick experiment to show that desaturation is one way to make the colours look more pleasant. I just went all the way to B&W. :stuck_out_tongue:

Actually, without any colour adjustments, I found the colours to be not great but not too bad. Here is a take with no colour adjustments. Excuse the noise.

@s7habo
I downloaded 20180922_FUJ8679.raf into one folder and your xmp into another, renaming it to identify what it was. Then I opened the raf in dt and loaded your xmp. I then went through each step, saving snips of the settings of each module. The next day, I followed the snips to work through it myself. Later that day, I opened the raf and tried loading your xmp again. It won’t load. I tried removing the raf from the collection and then opening it, but it still wouldn’t load.

I tried changing the name of the xmp back and copying it into the folder with the raf, and then opening the raf, but that didn’t work. I downloaded a copy of the xmp from my online backup, from which I could tell that my original copy had not been altered. Still didn’t work. I tried loading the xmp that I created when I walked through the steps, and that loaded.

This is the second time I have had this happen. The last time, though, I had changed the location of the folder holding the image and the the folder holding the xmp.

Do you or anyone else have any idea why this is happening?

The XMP needs to stay in the same folder as the RAF file. Darktable looks for an XMP with the same base file name as the RAF. Also the “master” copy of the XMP file is in darktable’s database; when a module is altered, a new XMP file is written to disk.

Also there is a checkbox in the preferences to look for a new XMP and darktable will give you an option to import it if it does see a new version.