Well, playing around with sigmoid was a really happy surprise. I’m finally closer to the colors I strive for. It also looks more tonally right to me (very subjective, of course). With filmic, not trying to preserve chrominance made it a lot better, but it was still more of a struggle than a love-at-first-sight.
I have gone through 50 pictures or so and I’ve yet to find a picture that looks better with filmic than with sigmoid. The difference between sigmoid and filmic (with preserve chrominance: no) was usually subtle, but filmic was quite often subtly wrong. Happy to get rid of the strange reds.
Personally, I don’t see how sigmoid couldn’t be the new default tone mapper. It could possibly save us from first time users opening darktable and quickly closing it again. Those who prefer filmic will likely be able to set it as their default.
I think the only thing I’m missing now is a better way to grade. I’m finding color balance rgb very limited for that purpose. I believe color eq would solve this. Until then i will continue to use cube LUTs after sigmoid. The g’mic film LUTs looked very good with sigmoid.
Looking forward to use sigmoid more!
EDIT: Almost forgot… Thank you for the hard work @jandren!
CB is fine for a lot and it really does a nice job, esp for global and tonal adjustments, but DT could still use a modern color zones module for tweaking hue vs hue hue vs chroma in a way that doesn’t require multiple instances and masks… I know AP started on a bit of a project for a tool like that but I don’t think it got too far…
First impressions of sigmoid compared to filmic. The images I tested it on are challenging because the subject is in shadow and background is sunlit. Sigmoid seemed to do a nice job revealing the color in the shadows while filmic seemed possibly superior in the brightly lit sky of the background. I had to recruit the tone equalizer module to balance the shadow highlights. Ultimately I chose the Sigmoid version for export with this shot. Is my observations surprising or typical of sigmoid?
I have now edited the post to include a sample of the scene.
Shared experience is good but without sharing images /examples about what we are experiencing it is going to be hard for @Jandren to comment or anyone else…
Contrast controls the slope of the S-curve, while exposure controls the crossover point, those are the two main parameters. The others ones are secondary to fine tune the results.
For that reason I think there should be an exposure compensation slider directly in sigmoid, because currently one of the main parameter lives in another module. As I understand it’s just a matter of multiplying the values by a constant before the sigmoid curve, but maybe there’s some extra complications I’m not aware of.
There is a very good reason for not having a “exposure slider” in the display transform. The whole scene referred workflow revolves around normalizing middle grey as 0.1845 instead of an arbitrary maximum value to one. This is why the sigmoid module works so well with the already existing scene referred modules! The whole sigmoid curve is actually pinned downed by the observation that middle grey (0.1845) should pass through this module unmodified. All other values on the curve are defined relative to that condition.
There was a slider for this in filmic, “middle grey” but it has been hidden for the same reason as it’s not included in the sigmoid module.
Yeah indeed that’s an issue. Since this is purely a usability issue (having related parameters in different modules) it should be addressed by the UI rather than changing the pipeline, and you can already open both modules (although it’s not super intuitive how), or maybe have them in the quick-access panel. That said I’m not totally convinced that having the option wouldn’t help in some cases, specially since changing exposure can be problematic with you have masks and stuff. I can put an exposure module before sigmoid to test though.
You should put an exposure module before it… There is a reason it’s there by default .
It’s also one of the things you should set before doing anything else… although I do realize it may be hard to judge if you set it ‘right’ from the start.
Set the contrast in sigmoid to something very high like 5.0 - 10.0 (use keys to type it in as its outside of the soft bounds)
You should now have a very contrasty image with mostly black and white stuff.
Adjust your exposure such that your subject (face etc) is right in the sweet spot between white and black.
Reset the sigmoid contrast value and work from there.
Helped me to get a grasp on where middle grey is.
Any further adjustments is preferably added in new modules so that you always can go back to that initial normalization.
In my opinion, this is the number one issue folks have either filmic or sigmoid. Determining the middle gray when you have a large dynamic range image is really important and mostly an artistic choice.
Just out of curiosity, how did you generate these images, did you use darktable with a test image with a linear gradient or did you do it different? If the first is the case, how did you do it, which image format did you use and why, how did you ensure that this survives colorin, etc.
It’s just a random image I had, probably downloaded somewhere, thus I wouldn’t read too much into it. For example you can see the “gray area” increase when I boost the exposure, but I’m not sure if that’s due to that particular gradient or to sigmoid.
I have been using the exposure auto picker. I leave it at 50 to see where it lands Often that is not actually bad. I guess you could refine that and pick a region that you want to be middle grey. With filmic and most of the images I edited 60 % for the overall image was a pretty reliable first guess. I also always started with filmic off so I could see how it visually impacted the image when added . From there filmic auto pickers were usually fine. I have not used Sigmoid enough to know if there is a sweet spot like that is.
That’s what I do too. I actually enable my cube LUT after sigmoid too when I set exposure. After that I dial down the contrast until everything looks fine and then turn off the LUT to see that the image is mostly within gammut before I enable the LUT again.
.
.