Done, I guess ?
Thanks Aurelian, this is excellent.
My question - which modules hard clip in the pipe (0-255) before filmic, and which modules don’t hard clip (0-infinity)? Knowing this will help understand which modules it is ok to push (because the detail can be brought back in filmic) and which modules it isn’t.
Modules I am particularly interested in:
White balance, Channel mixer, exposure, tone eq, color balance, contrast eq, sharpen.
“flatter”, I think.
On a different note, I always wondered whether filmic actually was a sort of level tool in disguise. And if it was, why filmic was not presented like a traditional levels tool, with draggable boundaries on a histogram?
Or perhaps like Capture One’s levels, where boundaries can be dragged both in, and out beyond [0,1].
Regardless, the scene-referred workflow is amazing; I am currently recreating old Lightroom edits in Darktable, and quite a lot of highlight recovery fumbling with levels and tone curves and highlights/whites in Lightroom becomes dead simple in filmic. Great work!
1 Like
Like this? (Current dev version, one of the views available.)
3 Likes
Note that clipping is much harder to see in that view. The white/black relative exposure sliders are the same as in your ‘levels’ tool, no matter which visualisation you use.
1 Like
Yes, I agree with @bastibe, perhaps replacing the two sliders white/black (or maybe also the mid grey) with draggable handles in the above chart would be far more intuitive
Very informative. Thanks. Will you put a demo on Youtube?
For me Filmic4 is a big improvement. Thanks Aurelien for once again a great article here about Filmic.
What I’m struggling with : I adjust exposure. In my case usually to +1 - +1,5.
Toggle under/overexposure and adjust white relative exposure and Black relative exposure in Filmic till clipping is (almost) gone.
Adjust highlights to my needs in Filmic.
Then Color balance : saturation and contrast. Afterwards contrast equalizer clarity (in most cases reduced to 40%).
Now toggling under/overexposure, clipping returns, so one is tempted to go back to Filmic to adjust white and Black relative exposure again which lowers contrast.
I tend to go back then to Color Balance to rise the midtones. Sometimes followed by the tone equalizer to rise general brightness.
Reading Aureliens explanation here, I’m starting to wonder if I’m too worried about the under/overexposure clipping indicators here ?
Are those indicators in dt still up to date or even relevant to the new workflow ?
After all that I can’t resist playing with a Lab curve to ad contrast, lower highlights… Following the introduction of scène-referred workflow this isn’t the way to go I understood.
Thanks in advance.
1 Like
The over/underexposure indicators are applied to the output of the whole pipeline. See Channel mixer behaviour (and interaction with filmic) in darktable 3.2 - #5 by elstoc
I think you dance around too much. You should adjust exposure to get your midtones right.
Adjusting colour balance will adjust filmic’s input, so avoid large changes there. However, if you change contrast and midtones in colour balance, you can also adjust highlights (which is linear scaling, like exposure) there, too. Also, why use the tone EQ for ‘general brightness’ – the whole idea is that you adjust specific areas independently (at least if you use detail preservation).
For colour balance: create a black-to-white transition in the Gimp (even better: posterise it in the Gimp, so you get steps that show up distinctly in the histogram); load it into darktable, and see what colour balance’s highlights (gain), mid-tone (gamma) and shadow (lift) controls do.
For contrast EQ, I use a trick of applying a parametric mask concentrating on the mid-tones; that way, highlights and shadows are usually not altered that much. Also, I leave a bit of safety margin in filmic.
1 Like
So, you’d want an ‘input’ range from, say, -15 to +10 EV, and move all the three points around, and then you’d switch over to the curve to see if it clips? What you can do now is to turn on the mid-grey slider in options, and voila:
Changing the mid-point is equivalent to changing the exposure, but filmic also adjusts the white and black point’s relative exposures, to keep them at the same absolute value. Halving the mid-point means increasing exposure by 1 EV, so the white point is now 2.9 + 1 = 3.9 EV above the new, lower mid-grey, and black is adjusted the same way. Doubling the mid-grey value means reducing exposure by 1 EV, and the white and black points are adjusted the accordingly (for white: 2.9 - 1 = 1.9 EV).
Though, @anon41087856, shouldn’t black move also in the same direction, not opposite, to keep the same value (and the same dynamic range)?
2 Likes
Thanks for good and clear explanation. I think I understand now more.
I’m struggling in my workflow to get the middle gray right as I don’t “see” it and it leads to jumping between exposure and filmic. I remember there was some discussion about adding the exposure also to filmic, but I agree it would just duplicate the functionality. This goes little off-topic from filmic, but would it make sense to add to exposure a color picker, that would adjust the exposure of the selected area to average middle gray? That would at least help me to get the starting point right. Or do have any other suggestions?
1 Like
It’s already there: adjusting the mid-grey point performs the same multiplication as changing exposure does. Don’t get too fixated on ‘mid-grey’, think ‘mid-tones’, like people, plants, the blue sky (away from the sun) etc. E.g. there are photos in this article showing which areas are the same lightness/brightness as mid-grey: https://mastinlabs.com/blogs/photoism/understanding-middle-gray-and-how-to-find-it-anywhere
2 Likes
The levels module is grounded into the display-referred range. That is convenient for the GUI because it is bounded in 0-100%. With filmic, you have no bounds, and the range is -infinity EV to +infinity EV. That would require to rescale the GUI space as you change the bounds, and you couldn’t add + 3 EV straight, for example, since the +3 EV would be out of the widget entirely.
Then, the histogram should be there in log space, therefore using an internal method instead of the general pipeline histogram, so the GUI thread (CPU) should fetch and wait for the histogram thread (CPU or GPU), and… That tiny little feature will be worth 1 month of debugging to sync the threads.
None of them clips anything except for the channel mixer.
You need to remember that the under/over-exposure preview doesn’t show where pixels clip, but only the pixels that belong to the top and bottom 1% or 2% of the histogram, depending on the threshold you set. So there are valid and clipped pixels alike, in there. And indeed, there is no reason to get alarmed if small regions show up there, unless you have like a quarter of your image that ends up in those 2% extreme.
Ok, I need to address that in the FAQ.
In practice, when you need to lower the scene mid grey, it means that you are working with HDR scene, so there is no point trying to keep the DR as-is and only sliding it.
Middle grey is the color of the grey GUI for darktable, and also the color of the background when using the color assessment mode. That is supposed to give you a visual reference.
2 Likes
Did you know that you could adjust exposure simply by placing the mouse pointer over the middle or right-hand-side of the histogram, and scrolling? Adjustments are made in steps 0.15 EV. On the left side, you can adjust the black level correction in steps of 0.001. “Histogram” in the darktable usermanual
Also, you can use dynamic shortcuts that allow you to change exposure from anywhere. “Shortcuts” in the darktable usermanual
3 Likes
Thank you very much for the explanation, Aurélien !
So I Geuss trying to get rid of old habits focusing too much on under/overexposure-indicators is the clue here. I’ll keep practicing 
@T_Yik Thomas Aurelien has several videos on his YT channel https://www.youtube.com/c/AurélienPIERREPhoto/videos
Thanks for the link. I need to learn to see the luminosity and not the colors. The link is good reference.
In theory yes, but in practice modern cameras have a DR of about 14 EV and the RAW file formats are anyway limited to 16 bpc.
A histogram with 16EV soft limits would require no resize of the GUI for 99.9% of the cases
Have a look at Capture One’s levels. Instead of a movable fixed lower and upper boundary, their boundaries have two control points:
(screenshot from their website)
Moving the lower points acts like a normal levels boundary. Moving the upper points inwards expands the output dynamic range. I think. This is not unlike your DR mapping view I think, but directly editable in the graph.
But I understand your point with regards to implementation difficulty of the log histogram.
Just a dumb question regarding the auto-adjustment buttons in the “scene” part of Filmic module.
Why “white relative exposure” is not set at a value related to a brightest point in a picture when clicking on the auto icon? Similarly, why “black relative exposure” is not set at a value related to a darkest point? I know that it would be useful mainly for high contract photos, but my personal feeling is that at this moment the auto icons are not very helpful and such a change would be more understandable…