Thanks. Tricky stuff!
The in camera settings only impact the out of camera jpg…nothing impact the raw from the camera wrt color space
1 Like
So your post just reinforces my comment…why can you call this a “filmic” edit…its your filmic edit with some interesting choices…taking your edit and doing nothing but resetting filmic…gives this…not the ghost that you provided… these sorts of comparisons are really not that useful in my opinion…
2 Likes
It’s just filmic with raised exposure
5DSR2210.cr2.xmp (7.0 KB)
Edit: the culprit is middle-gray luminance slider
1 Like
What I’m showing here is a series of colour-picker values from the two areas shown on the image.
The first pair is without exposure and filmic; the 2nd is with exposure raised; the third with filmic v6 (reset to defaults):
As you can see, the H values were restored by filmic (rounding to the closest integer, 30 and 36 became 32, 59 after exposure, and 31, 35 after filmic), and saturation was dropped.
LCh tells a different story, though:
The darker tone started from h=70 and went to 73 after exposure, then back to 71 after filmic. The brighter one, however, went from 77 to 97; I guess that was clipping, showing the ‘rat-piss yellow’:
With filmic enabled, the darker tone went from 73 to 71, but the brighter one fell to 68 (using max RGB).
Setting chroma preservation to no puts it at 66, luminance Y to 62, power norm to 67, Eucledian norm to 62, legacy Eucledian to 70:
Adjusting the white relative exposure to retain the highlights:
Maybe this (all but legacy Eucledian norm shifting the hue) would be worth reporting as an issue.
1 Like
5DSR2210.cr2.xmp (7.0 KB)
I think this is the best I could do with filmic only (and I really mean filmic only, I did not even adjust exposure), without going obscenely out of gamut. I used v5, for better results in this scenario.
1 Like
Way better, the neutral raw looks slightly more yellow but without a direct comparison this is good enough to me!
I should perhaps add that I immediately reacted to your interpretation of the scene quoted below when I scrolled through this thread. It stood as the best of the bunch. It looked “filmic”, in the original sense of the word.
Is PhotoFlow still being developed? The latest build I can find on GitHub is from 2020-08-04. What is this “experimental rgb hue preserving tonemapping” you’re speaking of?
This is going off-topic. Perhaps there could be a separate thread about that tone-mapping.
Yellow is a combination of red and green. Each channel individually goes from black > colour (eg. Black > fully saturated Red), but the effect of the three channels combined is black > white. If you turn off all your modules and look at the waveform (individual channels) you will see a small amount of information already near peak in both red and green channels. Individually, this tell us these channels are very colourful, but together it tells us they are also very bright. This gives you very little room to move in editing. If you boost exposure to make the flowers as a whole brighter, those peaks are suddenly way off the map, thus out of gamut (you can double check this by viewing the perceptual gamut scope instead of waveform. Boost exposure and watch those colours go out of gamut). Your gamut mapping (if not done in darktable, is done your monitor) will then put them back in gamut according to the rendering intent. In this case, and often with sunsets, we get the infamous rat piss yellow - which looks colourful, but garish and inaccurate. When you turn on filmic, which compresses the range and has gamut mapping, it will bring everything back in gamut, but either at the expense of hue (thus our yellow might become orange or pinkish as some say), or colourfulness (thus our yellow becomes desaturated and closer to white). How to avoid? You can avoid boosting exposure. The image as a whole will be too dark, but your yellows will be nice. If doing this, you can use base curve (or some custom rgb curves) instead of filmic, and boost the midtones, which will leave the bright yellow’s pretty nice, but have a different feel in terms of contrast. @age edits retain the yellow nicely because of his use of custom curves. Another option is to export two versions - one for the picture as a whole, and one for the highlights - and combine them in gimp or krita.
EDIT: If using single version of filmic for your edit, tone equaliser will be a big help. I didn’t push it far enough in my edit.
2 Likes
darktable (on the left) vs. darktable output, a TIFF in Rec2020, in a colour-managed viewer (on the right):
3 Likes
The tiff looks great!
So, it seems that for now, I’ll export Rec2020 TIFFs, then do
tificc -e Rec2020-input.tif sRGB-output.tif, and then convert to JPG using imagemagick, unless I can convince IM to do the conversion itself.
2 Likes
Interestingly, if I add the Rec2020 TIFF exported from darktable to darktable, it looks good in the darkroom. Top: raw file; bottom: exported+imported TIFF
Edit: opened https://github.com/darktable-org/darktable/issues/12175
3 Likes
Its one of those images with some extreme light and so how much you keep or how you hand WB might strongly impact the end result but I didn’t really do an edit or adjust filmic at all I merely pointed out that the edit shown for filmic was a long way off. I just took the edit presented and used default filmic settings instead with no adjustments at all so its not an edit just meant to show this notion of calling an edit a filmic edit…@age pointed out that it was a strongly modified custom grey value, the filmic curve was a bit clipped at both ends and the latitude was zero…
Your point is of course true, settings matter and make comparisons difficult. However your example is a very dark edit that avoids the difficulty of the photo. The blacks look crushed or at least way to dark. (On my phone so can’t say for sure)
The question is does the tonemapping allow you to map the tones and achieve correct and pleasing results in a controlled way. I’m sure you can but it’s probably possible to make it more robust for users.
For sure and it wasn’t actually an attempt at an edit. I was merely trying to show that one example presented for filmic which appeared quite like a ghost to me was not indicative of what using filmic as a tonemapper for this image was likely to produce or even capable of…in the image I just changed the edit of @age to have filmic at its defaults. @age had not added any exposure preferring to try and tweak the image with the middle gray slider instead so for sure me just flipping it to defaults darkened the image but it wasn’t imo that far off the presented examples and a subsequent edit by @mikae1 was also quite acceptable.
Perhaps the question is not if you can but more so how hard is it to achieve consistent predictable and desirable results wrt another method. I think many believe this is where the sigmoid module has shown some promise??
Thats certainly how I see things. With the caveat that I wonder if the pinkish results are in any way correct or acceptable. It’s such a nasty effect that a good tool/algo should avoid it imho but perhaps I’m missing some situation where it’s desirable.
edit: just to add that the pinkish thing is particularly noticeable with portraits. We’ve seen quite a few threads with people struggling with it.
A real hdr image from a Sony hdr video demo (4000 nits peak brightness),I have extract a frame and done the conversion to rec2020 floating point :
Sony 4K HDR Demo - Camping in Nature.mp4.7z (10.8 MB)
the frame
My version (actually it is sigmoid )
hdrframe.pfi (16.1 KB)
RGB per channel tonemapping
Darktable no preserve chrominance
Darktable preserve chrominance
Sigmoid with film-like hue preserve (the same used in rawtherapee)
2 Likes
Export from darktable in Rec2020, reload it, export as sRGB without any changes - how does that look?
1 Like