Some input on "recovering skies"

[Dinobe]

I’m looking for assistance and input on how to recover the skies in this shot. This image was taken years ago and had I processed in Lightroom. While stuck at home and as a kind of training in Darktable, I tried reprocessing this image.
It feels like I’ve been a in fight with Darktable the last couple of hours and I can’t figure out how to darken the skies/recover the skies.

I’ve tried:

• adding an Exposure module, -2.5 EV, with a gradient over the sky → clouds turn magenta and purple
• adding a colour balance modue, dropping shadows, midtones, highlights to -50% (the minimum) → some blue clouds reappear
• adding a tone curve, dropping everything to a minimum
• adding a RGB curve, dropping everything basically to 0, → some magenta clouds reappear
• adding filmic, set extreme luminance saturation to 0% → some ugly plain grey clouds reappear
• tried adding a tone equaliser over the sky and drop everything, but couldn’t get clouds back

After 3 hours of trying I’m giving up. What am I missing? I feel like I’m fighting a piece of software and not working with one to achieve a goal.
After every step I feel like: why have you done that? I have no problems learning something, but this is becoming ridiculous. There must be something key I’m just overlooking.

You can see the Lightroom Export, the current export from Darktable. The foreground looks good / similar to what I had. The clouds are nowhere near, and the information is there as LR can recover them. The shot might not be perfect nor the processing might not be your taste, but I’d just like to know how to achieve those clouds.

Darktable export (similar result after tone cure, rgb curve)

This file is licensed
(Creative Commons, By-Attribution, Share-Alike)
IMG_1324.CR2 (22.6 MB) IMG_1324.CR2.xmp (21.3 KB) Darktable export (filmic)

This file is licensed
(Creative Commons, By-Attribution, Share-Alike)
Darktable export result:

This file is licensed
(Creative Commons, By-Attribution, Share-Alike)
Lightroom result:

This file is licensed
(Creative Commons, By-Attribution, Share-Alike)

I 'm releasing these images as (Creative Commons, By-Attribution, Share-Alike)

I also notice some ugly grid pattern in the Darktable exports, apparently when switching on "high quality resampling " on export, I get these grid patterns…

[elGordo (Gord)]

Thanks for posting the images. You should give them a license so that folks here can work with them and provide suggestions. See PlayRaw stuff to keep in mind

Also:

• What release of darktable, obtained from where?
• What operating system are you running, and what release is it?

[Dinobe]

I’m running Darktable 3.0.2 (github) running on Windows 10, 1909

[patdavid]

@Dinobe would you mind terribly explicitly stating the license so we can look at it for you?

[Dinobe]

Sorry, struggling on how to do that exactly

[patdavid]

Just post that you are releasing the image under license XYZ (preferably Creative Commons Share Alike).

[Dinobe]

Hope I got it right now, still learning…

[CarVac]

So in this photo you have clipped two of the raw channels, leaving only the red channel left to retain any highlight data.

To deal with this, you need to change the highlight reconstruction module’s method; “reconstruct in LCh” works well for this image. Then reduce the clipping threshold.

I do have a question, though: was the sky actually blue? To me it looks like entirely overcast weather, throwing Lightroom’s output into question.

[Dinobe]

@CarVac

I do have a question, though: was the sky actually blue? To me it looks like entirely overcast weather, throwing Lightroom’s output into question.

No I don’t think the sky was blue, must have been an artistic interpretation I gave to the sky in LR . What was I thinking so many years ago…

I’ll have a look at that reconstruct in Lch option

[baongoc124 (Bao)]

@Dinobe I played with your image a little using same method as @CarVac with addition of color reconstruction module.
Here is the result with darktable 3.0.1

IMG_1324.CR2.xmp (11.1 KB)

[CarVac]

Here’s a literally-zero-effort output from RawTherapee using the Color Propagation highlight reconstruction method and then slightly reduced exposure compensation, just to see what you have to work with:

And here’s what I did in Filmulator, which has the same backend Color Propagation highlight reconstruction:

• Auto CA correction 1 (on, in previous versions)
• Highlight Recovery 2
• Exposure Compensation -0.67
• Shadow Rolloff Point 0.034
• Drama 76
• White Clipping Point 0.422

[Daniel_Catalina]

My approach with ART:
IMG_1324.jpg.out.arp (11.9 KB)

[Nathan_Crabtree]

With DT

[peterdb (peter)]

dt

IMG_1324.CR2.xmp (21.4 KB)

[Dinobe]

Ok guys, thanks a lot for pointing me to the ‘reconstruct in Lch’

I steared away from it as it said “Manually adjust the clipping threshold against magenta highlights (you shouldn’t ever touch this)”

I had to drop the slider to 0.788

I’m getting much better results. I think the image could still some local dodge and burn. But it’s already looking a lot better.

[MStraeten (MartinSt)]

i found 7D needs a lower raw white point. I found 13580 a better white point reducing magenta in clipped highlights very well. Additional hightlight reconstruction in LCH an a one-size-fits-it-all filmic style (see Filmic RGB - loss of detail and saturation in high contrast scenes - #7 by MStraeten) gives a good base to tweak details - maybe desaturate the sky a bit to reduce the color difference between clipped and unclipped areas, increase color contrast in the green areas, …

IMG_1324.CR2.xmp (6.3 KB) (the file was processed with filmic v4 module but v3 setting, so there might be errors when importing the xmp with 3.0.x)

a little bit tweaking:

IMG_1324.CR2.xmp (8.8 KB)

[heckflosse (Ingo Weyrich)]

@MStraeten

i found 7D needs a lower raw white point. I found 13580 a better white point reducing magenta in clipped highlights very well.

From RawTherapee camconst.json:

    { // Quality A, ISO and aperture WL data by CharlyW at RawTherapee forums, missing samples safely guessed
        "make_model": "Canon EOS 7D",
        "dcraw_matrix": [ 5962,-171,-732,-4189,12307,2099,-911,1981,6304 ], // Colin Walker
        //"dcraw_matrix": [ 6844,-996,-856,-3876,11761,2396,-593,1772,6198 ], // dcraw
        "ranges": {
            "white": [
                { "iso": [ 100, 125 ], "levels": 13480 }, // typical 13584
                { "iso": [ 160, 320, 640, 1250, 2500 ], "levels": 12550 }, // typical 12650
                { "iso": [ 200, 250, 400, 500, 800, 1000, 1600, 2000, 3200, 4000, 5000, 6400, 12800 ], "levels": 15200 } // typical 15304
            ],
            "white_max": 16383,
            "aperture_scaling": [
                // no scale factors known for f/1.2 and f/1.0 (no lenses to test with), but the
                // typical 12650 white levels maxes out at "white_max" for f/1.4 and below anyway.
                { "aperture": 1.4, "scale_factor": 1.250 }, // guessed
                { "aperture": 1.6, "scale_factor": 1.150 }, // guessed
                { "aperture": 1.8, "scale_factor": 1.110 }, // 15196/13584
                { "aperture": 2.0, "scale_factor": 1.080 }, // 14734/13584
                { "aperture": 2.2, "scale_factor": 1.050 }, // 14386/13584
                { "aperture": 2.5, "scale_factor": 1.040 }, // 14272/13584
                { "aperture": 2.8, "scale_factor": 1.030 }, // 14042/13584
                { "aperture": 3.2, "scale_factor": 1.015 }, // guessed
                { "aperture": 3.5, "scale_factor": 1.000 } // guessed negligible
            ]
        }
    },

[Soupy (Tim)]

Tricky one. The highlights were very recoverable, but the fringing atop the hill was a challenge. Would have been better to do two different versions - one for highlights, another for the rest - but here it is in one. Using darktable.

IMG_1324.CR2.xmp (25.3 KB)

[Dinobe]

@MStraeten

i found 7D needs a lower raw white point. I found 13580 a better white point reducing magenta in clipped highlights very well.

You lost me there, but I’ll have a look into that…

[MStraeten (MartinSt)]

Its in module raw black/white point one of the default modules and sets some basic values for the useable range between min black and max white level of the sensor. “Modules” in the darktable usermanual