I’ll probably add presets with auto on and off, so that’s a detail only, no actual coding is needed. Having it in the common/basic section requires a bit of coding, to keep sliders in sync for the 3-tab layout.
Thanks for the feedback (and for voting)!
@kofa , this is really good letting people express their preferences and taking them into account in the development. This could be a template for future modules?
I like that filmic can do this. A straight line on the right can give a better highlights look as opposed to “rolling off”.
Not really, it all depends on what is to be decided. For a lot of cases, the GUI layout is dictated by the internals of the module, something the average user wouldn’t know about (and probably doesn’t need to care about).
And, this kind of forums only reaches a limited number of users (even members here don’t all check in every day), who may not be representative for all users…
Not sure that’s always a bad thing. If your image has a low dynamic range, you don’t need any roll-off, so forcing an S-shape in that case would mean too much contrast in the mid-tones, and less in the shadows and highlights.
Good for most situations, but I am regularly dealing with photo’s taken through a microscope, where I had exactly this situation : very low dynamic range, with relevant details in both the “shadows” and the “highlights” (you don’t really have shadows and highlights here.)
For such images, you may not even need a tone mapper (I assume for scientific imagery, one may simply want to maximise contrast to see the details). I think @priort has posted several PlayRaw contributions where he omitted any tone mapper and used other tools.
@rvietor , 65 users have contributed to this thread, and with the 20k views, I imagine quite a lot more have read about AgX here and have had the chance to contribute if they wished. I think this represents valuable user input. Let’s bear in mind that changes are sometimes made to DT without any airing at all here in pixls. Also I think the pixls platform/software is a nicer way to have a discussion than github.
Where you say “members here” do you mean the developers? They can of course check in to github and pixls as often as they like or need.
I appreciate there are balances here. Pixls is good for user involvement and discussion; github is necessary for managing development and the software components. Users have wishes, ideas and criticisms but in the main can’t or don’t contribute their time to development (though some contribute financially of course); developers do things for their own reasons (just like the rest of us) so are free to pursue pet projects without a big discussion in pixls.
You mention the “average user” - is there such a thing?!
I’m glad that many have voiced their opinions and have contributed with observations, examples, questions and suggestions. Yet, many users, who frequent other communities (whether it’s https://www.reddit.com/r/DarkTable/, Flickr: The darktable Pool, Redirecting... or others) and for them signing up to yet another forum is not worth it. (Their loss, of course. :-)) We don’t know what they think. And even if every single darktable user chimed in and added their opinion, there is no way I could come up with a solution that suits everyone perfectly. I trust that those who are here will not have too much to complain about. 
I think it is also true that not all modules (e.g. deeply technical ones, where the internals of the algorithm, and not merely the UI have to be tweaked) can/should be developed that way. I was fortunate enough to be able to build on sound foundations: the work that went into implementing AgX for Blender, and concrete code in the rgb primaries module and the primaries functionality in sigmoid, which was also inspired by that work.
I wasn’t saying asking user preferences here was bad.
I was commenting on using that as “a template for future modules”.
But you mention 65 participants for this thread. You think that’s a significant percentage of the user base?
His point was that github would get even less engagement, so this is as good as it gets.
Afaik the only developer who tried to gather user opinions/stats on a larger scale was Aurélien with the yearly darktable questionnaire. Even that didn’t have a lot of data.
Now that I am a bit more comfortable with the controls I am editing more by eye and the basic controls offer more than enough for me. I am voting to keep it and tuck it away so that users that prefer otherwise can have access to it. But really I am happy to work with the module in any configuration. I think given its a single slider it could easily be mapped to a shortcut with the mouse …one of those ones with the scroll wheel or dragging the mouse up and down or across the screen like a virtual slider…
I have been a bit busy orienting all of our new summer staff in the lab so I have not been doing many edits over the last couple of weeks but I have now done enough to sort of see where my workflow tends to go for a starting point in most edits…
I do the standard lens correction, denoise and CA removal. I usually don’t tweak denoise too much but just the preserve shadows until I like the look or often I will only do the color noise preset and then tweak luma noise with either surface blur or astrodenoise…
Exposure with the autopicker…if I don’t like it tweak by hand or more often a second sample on an area of interest.
AGX
Rel B&W auto pickers
Pivot picker with gamma unchecked
Adjust that to overall taste…so for a global “toning”
Tweak contrast around the pivot…
In some cases tweak toe settings
I almost always finish with some tweaks in the look section as I find they can really add a nice final touch
From there for most edits will have local contrast, 1 or often 2 Diff or Sharpen instances and then tweaks to ToneEq, rgb CB and CC or ColorEq.
I must say that often I find the number of extra modules and the work that I am needing to do with them seems to be greatly reduced with the use of AGX… Having said that I do pretty simple edits compared to many others here…
I know this is more of a word-play, but AgX’s philosophy revolves around the idea that it converts exposure data that is not yet an image, into an image made for a specific type of medium.
In this philosophy, when ever you see an “image”, it’s always “formed” in some way. If you directly encode the exposure with the transfer function curve and send the exposure data to the screen, you are interpreting the curved exposure data as an image made for that specific display, and the image is formed at the point the signal reaches the hardware, where the signal is clipped. Which means, “no tone mapping” is also a kind of image formation.
Even if you turn off the demosaic, and directly display that, you are still actively making a choice to interprete the bayer pattern as an image by sending it to the screen. Basically, AgX’s philosophy emphasizes the idea of forming an image/picture.
It’s true that different kind of images may require different method of forming images, but if we are discussing AgX, I would like AgX’s philosophy to be brought along with it.
Whether the curve is strictly S shaped or not doesn’t really matter if the discussion is detached from what the actual image looks like. The curve is just a tool to author the visual grammar written in the image as a “text”, for the audience viewing the image to “read”.
Well, you don’t need a tone mapper, but filmic does give some advantages. E.g. the log transform has an effect, and you can compress the dynamic range.
But my point was that “losing the S-shape” isn’t always a bad thing.
Very impressive tool. It kind of feels like 5 moduls in one. I need to get used to it but it gives a lot of control. I wonder if the sliders do not interact with each other?
This echoes my thinking - we are editing images, not curves. The curve is a helpful guide, but not the end goal.
But if others find that having the curve displayed a particular way helps them make better images, then ok.
Yesterday @s7habo and I had a fruitful discussion, he taught me a few things about agx.
@priort: based on my previous experience with filmic, I also used to start with auto-picking the relative exposure bounds (relative black) white exposure). However, Boris showed me that first setting the pivot and contrast, then using the shoulder and toe power controls allows for better control of highlight and shadow contrast. He only touches those controls if he wants to fill the output dynamic range (the histogram/waveform/rgb parade), and if unable to do so using toe/shoulder power. For example, if he wants to reach white, but the shoulder power does not take him close enough, he may lower relative white exposure. Alternatively, one could use the shoulder/toe start (so the linear section is longer, and the curved shoulder/toe starts later; those can almost always drive the curve into the floor (y=0) or ceiling (y=1)). I’ll see if I can edit only using those sliders, and get good results. If so, the exposure controls may become ‘use in case of emergency only’ tools.
And he also showed me how auto-gamma/power (keep the pivot on the identity line) is not a good fit for agx. I’m thinking about removing the checkbox and corresponding code; even if I keep it, it’ll be disabled by default (if you want to keep it on by default, you can create a preset and auto-apply it). I now think I’ll keep the gamma slider, but will probably hardly ever touch it, based on what he showed me.
@kofa, @s7habo This is a wonderful tutorial. I tried the process and it gives excellent results. BTW, I used the curve as a visual representation of the dynamic range (but not the guiding principle). Also the Gamma was set at 2.2 and the identity line unchecked. The photos were all in daylight
I have to try this on LED and artificial lighting conditions and some other extreme lighting conditions. This will be my next experimentation
However, Boris showed me that first setting the pivot and contrast, then using the shoulder and toe power controls allows for better control of highlight and shadow contrast. He only touches those controls if he wants to fill the output dynamic range (the histogram/waveform/rgb parade), and if unable to do so using toe/shoulder power.
Sort of building the image from inside, around the subject and only then applying limits if needed. Makes a lot of sense to me.
And he also showed me how auto-gamma/power (keep the pivot on the identity line) is not a good fit for agx.
Curious to learn more about this. But I was already happy with the capabilities of agx before the checkbox came up.
Limits are always applied (the defaults are -10 to 6.5 EV sound mid grey).
The toe and shoulder power ‘straighten out’ the central part of the curve, contrast remains (almost) constant over a longer range, then drops abruptly.
Thanks for sharing that feedback from @s7habo, I was having the same feeling about that function but I wasn’t really sure if it was just the way I was using it…
I have a question: if you first adjust the pivot, contrast, toe/shoulder power, etc. without touching the relative exposure controls, but then you still aren’t using the full dynamic range available and you want to, does clicking the auto tune levels picker mess up your previous adjustments? Or does it largely retain the curve you have just created and just “stretch” the black and white points?
Based on my testing, that auto picker does seem to “undo” the adjustments you’ve done, which would suggest we should NOT use this picker with this suggested workflow. Instead, we should only use the relative exposure sliders individually. Am I right?