Same, I do like just to adjust it to taste when needed. There are also a lot of images where the preservation operation is unnoticeable, and it doesn’t matter what you put it to. What value did you end up with for this particular image?
and
These two observations about the loss of detail in highlights and better looking bright check nicely sum up the difficulty for a display transform. Those two requirements are competing with each other, and I think it is impossible to solve that problem without taking the local context into account. Neither filmic nor sigmoid does that; both are per-pixel-based in their operation. So you have two options to bring out more details in a bright area, either dodge it with a local edit or increase the local contrast.
This brings me to:
I think it would be even better to add the local contrast adjustment before the display transform module. D&S is the option at hand today, but a modernized local contrast had, in my eyes, been a very nice addition to darktable!
Nice to see your experiments with both normal daytime images, sunsets, and epic landscapes!
About a recommended workflow, I will try to use the discussion in this thread as a baseline for writing them later on. There are multiple methods to get to the same point, but two possibilities are the following:
-
Activate and forget. (Default settings, contrast=~1.5, skew = 0, per-channel, hue preserve=50 or 100%) Do all the editing in other modules. The big difference to the recommended filmic workflow is that you will need to actively use the color balance rgb contrast slider to tweak your contrast setting.
-
Sigmoid contrast as a start. Same settings as above, but after finding the mid-tones. Tune the sigmoid contrast setting until you are close to satisfied/“neutral” with the saturation for your mid-tones. Then work with other modules as you normally would. Adjust hue preserve to taste. Skew will be part of the equation but try to save it for later in the editing (my gut feeling here).
Those are my two most used approaches so far. There are more possibilities.
The short answer is no. Both assume that the input data is unbounded, and both give a bounded output. Sigmoid would, in this case, not become very white (converging to white requires large values on the input, larger than filmic ever will output). You may, however, apply both in parallel and then blend the two results together. Sadly not supported in darktable, but export both images and then blend them in Gimp or whatever. A node-based image editor could make crazy things like that doable.