Apologies for the multi-posting. But I am done with a first draft of my LMT for ACES 2.0.
My goal was to provide a pleasing and natural aspect out-of-the-box on all the images I could test on. As expressed a couple of times in this thread, I struggled mainly with the blues which are very hard to untangle.
But the workaround that I found was to apply the RGC after my LMT. So this LMT is made of two steps: a grading pass for the pleasing aspect and the RGC to “contain” the data. That´s a bit of duct taping but I could not find a better solution for now.
I will detail here the process of fabrication. Maybe it will inspire others to come up with their own LMT and also start a conversation about what “pleasing” means. I am sure there are better ways to describe and analyze pictures than “it is is all creative”.
Thanks for your attention.
I generally start with a hue sweep in ACEScg where you can observe (among other things):
- Due to its “chromaticity linear” nature, ACES 2.0 will mainly go “pink” as exposure increases with reds and slightly violet as blues go brighter.
- We may also look at how the yellows are outstanding and probably breaking g0 threshold (the brighter yellow patches look emissive).
- The ACEscg green primary row may also look too “desaturated/pale” (it should almost be like a laser, right ?).
- Finally, the blues are too dark, compared to its surrounding hues.
So, the LMT tries to improve those aspects. Mainly by reducing purity and gently bending the hue paths. I also added a contrast boost to sit somewhere between ACES 1.X and ACES 2.0.
But those patches do not tell the full story and are missing a key component: gradients.
So generally my second step is too look at ACEScg primaries´ gradients:
On a red ACEScg primary, I try improve mainly two aspects:
- With ACES 2.0, there is like a clip happening that prevents the volumetric light to look “smooth”.
- And also, we are missing a slight hue path bending to go slightly more towards orange.
But if you pay really attention, I could not unfortunately get rid of the “kink” entirely.
Let´s have a look now at a blue ACEScg primary:
We can make the same observations here: by default, we get a breakup in the light and some funky hue distortions towards magenta are happening. This example here was the most difficult to deal with. I had to desaturate quite a bit to improve on those aspects.
Once I am done with those simple examples, I generally start looking at more complex stimuli and keep refining as much as I can. Because something that is missing from the previous images are the relationship between hues and how they “blend” together. This is super important.
For example:
Although I think my LMT improves overall this image, there is one specific area that is slightly worse. If you look at the penultimate row between blue and violet, the transition is actually less smooth in mine.
But that is a compromise I had to make for this kind of pictures:
Hopefully, with my LMT, the blue light reads as “light” on the table and the transitions between the wood (on the border) and the bright and colorful lights are “smoother”. Although I reckon they are far from perfect.
Here are more successful examples from my “exercise”:
So here, the obvious improvement is the hue bending towards orange and yellow. But there is something else at stake here. If you focus on the relationship between the grey smoke and the explosion (especially at the bottom right of the simulation), I managed to invert it. Because ACES 2.0 was the only DRT (that I know of) to create this effect.
We may now look at the sunset that Charles was mentioning:
Same thinking here. The hue bending helps with the pleasing aspect and I also managed to mitigate the polarity issue in the sun area.
In the following example, we can observe that peppers and apples look more natural and pleasing with my LMT. You may also observe that the contrast boost helps to make the specular reflections “sparkle”. We will come back to that.
Something I noticed very late in the image above are the blue reflections from the bottle on the red and yellow pepper (screen left). I think my LMT also improves this area.
And since we are on the topic of blue lights reflections, we shall move to our next example:
I focused here a lot on the blue lights. With ACES 2.0, it looks like thee is a dark ring around them. You may also see them in the blue reflection in the window (right screen). My LMT tries to give them a better energy representation.
It is worth noting here that my LMT is a bit saturated. This is mainly because I come from a “feature animation” background and we generally tend to prefer saturated images.
Another image I spent hours looking at:
Several things worth pointing out here:
- The contrast boost helps overall the image because ACES 2.0 looks too “flat” by default (this is by design). I was really careful to not push contrast as much as ACES 1.X.
- A good area to focus on is the shoulder (that is in the sun) of the screen left lady. With ACES 1.X, the contrast is so high that you loose all shaping. And with ACES 2.0, the “chomaticity linear” natue of it does not give a pleasing skin tone.
- Somehow my LMT sits between those two. Hopefully, the sunny skin looks pleasing and the contrast boost helps without losing the shaping in the “highlights” region. If you focus on the face, you will hopefully see that the skin “sparkles”.
- Also note that the orange patterns on her shirt do not look emissive anymore with my LMT.
We can also compare more “extreme examples”:
Same overall thinking here. By gently tweaking the hue bending, the brightness and the purity, we recover some shaping on the faces and a pleasing aspect.
More examples about skin:
Hopefully the contrast boost helps overall and if you look carefully at the lady on the left´s face, you may see some improved skin sparkling on her. I tried to retrieve as best as I could the information but in a very subtle way.
Of course we may compare my favorite stress test:
You can see how I struggled in the blue region unfortunately. But I think the rest of the hues improved with my LMT.
I checked also my Lego characters to make sure yellows behaved as expected:
The three bricks screen-left use ACEScg primaries. The green one might go slightly too much towards yellow. There was a fine balance to find which was not easy.
Finally, I also spent quite some time on the Grinch images:
Few elements to point out:
- The contrast boost helps overall with the aspect, especially the snow.
- The shadows with my LMT look blue and not magenta. I think it helps a lot.
- The grinch´s colour looks closer to its original intention.
- The christmas sweater relationship between green and red is also closer to the intended look.
In conclusion:
- If you are reading this, thank you. I spent a lot of time on this LMT and I hope I was able to explain clearly (part of) the process.
- I focused on a very holistic approach here rather than focusing on specific issues. My goal was really to give ACES 2.0 a more robust starting point.
- This LMT will be available later this month on my Git as part of an OCIOv2.4 config.
- I intend to share it with the Blender community because I think it will give CG artists and students a better starting point.
- Although I agree my LMT is not a solution per-se nor perfect.
- I also acknowledge that this long block of text is missing some proper thinking on the “why” I think a picture formation is successful or not. That will be for another time.
Finally, this “LMT exercise” has triggered an interesting question for me, which I would like to share. If you look back at the first hue sweep with the squares, in which order do you guys think the primaries should reach “white” ? I tend to think reds first, then green and finally blue (roughly thinking of LMS here…) but I would love to hear some thoughts on that.
Thanks !