I wanted to delete this image for good, but figured it would be nice way for you to show me how powerful DT is in saving clipped highlights and recovering the detail.
I’m still new in DT and would like to see from the experienced people how good it can be for this task. Because my success in saving the detail just wasn’t good, looked artificial, while RT produced so much better result that could actually be used.
Please use DT only. Thanks.
_V3A0302.CR2 (40.8 MB)
License: Creative commons
Can you add a license please?
There isn’t much you can do when +/- 33% of the image is clipped…
I think one should forget the surroundings and focus solely on the two females:
I have to admit that I did not spend a lot of time on this one.
Both are done with (a hybrid) darktable 3.1.0
I did decide to spend a bit more time on the sky and pool for this one. Also means that the overall look is more that of a overcast, sombre day. Sky is very problematic to “fix” though… Anyway, this was done with a “normal” darktable 3.1.0:
Thank you very much Jade! Well that’s exactly the problem I faced before - floor bricks were just irreparable in DT! It looked like it was extracting detail from JPG and not RAW.
RT on the other hand, did wonderfully both the floor and a bit of the sky.
I prefer the sky on the first two images you did. It’s kind of bluish, while on the third image it looks stormy! 
I also like how you focused on the people on the 2nd image!
How did you get DT 3.1? I have 3.0.2. The difference in recovered parts is HUGE on DT 3.1 and on par with current RT.
Update: my DT 3.0.2 hangs when I try to load your xmp, probably due to 3.1 ver.
Very nice contrast and colours, thanks. Punchy. I like it a lot!
Still a bit washed out floor, but I guess that’s the limit for DT 3.0.2, the way it handles extreme overexposure?
We could crop the front part and focus on the people and have totally usable photo here.
Very nice editing too! A bit hdr-like.
Thanks guys, I learned a lot from your changes!
The latest filmic rgb (v4; not available in 3.0.2, but in 3.1.0) has an excellent reconstruction option. This works in union with the highlight reconstruction module (and also exposure). It can do things that are not possible in the current stable versions.
The original does seem to have a hint of clouds (hard to be sure due to the clipping), but I have to admit that it is rather hard to get that back and it does result in a darker/stormy/sombre end result.
3.1.0 is the latest development version and not yet available for the general public. If you are willing and able you can build it yourself (build guide on GitHub). It is updated daily and not always as stable as you hope for (it is development…).
In August the new, stable dt 3.2.0 will arrive which holds all the goodies that are already available in 3.1.0.
The xmps created with a newer darktable are not compatible with older version (and might hang dt when trying to load them), hence the mentioning of the version(s) that I used if not the latest stable 
I have darktable 3.1.0+2446~g7d3d982d4 (compiled from git development a day or so ago) and used filmic RGB 4.
First attempt was to let the recover highlights module work in LCR mode.
After that, I made a pristine copy and turned off the recover highlights module, letting filmic do all the work on highlights.
I wanted to make the focus of the image the people, so I added a tone equalizer (to lighten the subjects and darken the top and bottom, more or less) and a duplicated exposure with a feathered, fuzzy, inverted circle mask over the area with the people (so that everything else becomes darker).
However, that affected the data filmic got, so its highlight handling let in too much semi-clipped data. Magenta invaded.
I then dialed back the highlight color, but it took away a little too much color in some of the non-sky (and sky reflection) color in the image. So I moved the highlights color back up and got a magenta cast in the sky (and reflection) again.
I then compensated for the magenta by “painting” a blueish color back using the colorize module with a parametic mask, to only affect the magenta color and then added two lasso masks to generally select the sky and the pool. Of course, then I feathered the mask.
Yes, painting the color on the sky is cheating a bit 
, but it’s done at the end of the pipeline and it it looks natural enough. I don’t know if the sky was completely grey, a touch blue, more blue, or some shade of magenta (due to a sunset) in real life. I’ve opted for a slight desaturated blue, which is most likely.
I went back to the image a third time to create a simplified version where I don’t add any additional modules in the scene-referred mode (official mode in darktable > 3.1.x, which is filmic-powered), yet only have filmic handle the highlights (again), to show a mid-step, and got this:
_V3A0302_02.CR2.xmp (5.9 KB)
It’s a little different from where I started with the middle photo, as the parameters have changed. However, as it doesn’t have the tone equalizer sandwiched in-between exposure and filmic, it lets filmic handle the extremes better, and doesn’t need the compensation, like the one in the middle.
Perhaps reordering the modules might make sense for this photo, to let filmic handle the highlights better and still be able to adjust the tons of the image with the tone equalizer.
Everything’s a tradeoff, I suppose?
So here’s one last one, where I built on the previous simple version, letting filmic handle the highlights itself, but reordered the modules to stack tone equalizer on top of filmic, then added a second exposure on top of that (to slightly darken everything but the subjects):
Edit: After seeing the image here, I’d probably further work on it, by making a mask for the equalizer to let the pool reflection shine through and then add another exposure module on top with a mask to slightly darken the chair highlights.
But, overall: This was a fun challenge and it was interesting to try out filmic 4’s highlight recovery.
@maboleth
Now you got some results from DT users, do you mind if I post result from other SW as it is an interesting exercise? or is it too soon?
Another attempt trying manual highlight reconstruction using the channel mixer module.
No, of course, go ahead. I’m mostly here to learn more about DT and how people fix/edit things means a lot!
Thanks a lot! You really tried so many ideas.
I think I like 3rd one best. It looks like a good tradeoff between nice contrast and resurrected whites.
I had a little fun with this one. How about this:
Not entirely fair though, this being a composite of 2 images 
This is your image with part of a sky/clouds from one of my own shots and it was done using the watermark module.
The watermark module takes an svg as input and darktable does not export to svg so I had to use inkscape to convert to svg, which is the ‘small step’ that isn’t done with darktable.
Nice. Even the reflection in the water looks like it was covered with the same type of clouds you added.
I’m so happy that DT 3.2 would offer much better tonal reconstruction of the overexposed floor tiles! Can’t wait for the final release. Rumored 1month from now.
3.2 is scheduled to be released August 10.
Experience learned that these planned dates aren’t written in stone though, it might be a bit later/earlier.
My take with ART
Log tone mapping, highlight reconstruction/color propagation, local contrast, color correction of sky.
Replacing the sky as done by @Jade_NL is the solution when sky is completely white to save a photo.
Thanks! Current (a)RT has the best highlight reconstruction of all! Floor tiles look totally nice.
Yeah, when sky is dead flat, you can do what Jade did with 2nd image or do some slight bluish cast as he also did on the first image. Anyway, very nice!
Sorry another non Darktable post. I found that Rawtherapee always had the best highlight reconstruction ever since I started using it in 2007. At that time it was far better than any commercial software.
In ART you can also apply the dehaze module specifically to the highlights and this emphasises any details in the highlights. It is good for bringing out textures in clouds.
On my phone now so don’t have examples.
