Hi, I revisited the blenderartists thread and saw a lot of activity there. So, here’s an updated AgX for ART based on the latest version by Eary_Chow. (To use it, either uncompress it or change the extension to .clfz)
AgX - EaryChow.clf.gz (1.4 MB)
6 Likes
For those interested, there’s now an open pull request to replace Filmic (as in Blender’s current picture formation method, not to be confused with darktable’s filmic) with AgX in Blender.
Thanks for the info! If somebody is interested, I can easily generate the LUTs that can be used in ART.
1 Like
I’m interested!
@flannelhead do you know where I can find more info about how the guard rail works? I only see a LUT in the repository, but I’d be interested in the details of the algorithm. Thanks in advance!
Maybe this could help.
A brief description
The code
2 Likes
Yes, I would’ve also posted these.
As I have followed the thread quite closely and took part in the guard rail experimentation, I can probably answer at least some questions if such arise from reading the links @agriggio
Awesome, thanks to both! I will read the material and get back with questions. Basically I’m trying to understand whether the guard rail can support HDR screens (say 1000 nits). I quite like how it handles the “path to white”, but the AgX config seems to be limited to SDR as far as I understood, which is a pity. (If I’m wrong, even better 
In this post last year, you release another LUT with the same name, but it was much lighter (<19 kB) than this one, and of course the result is different.
Can you elaborate on the differences and use cases for both?
This is for the latest version, a lot has changed in the agx formula in the meantime. I didn’t follow the whole development but I think the last one is supposed to supersede the previous…
1 Like
The more recent version includes the guard rail transform which is AFAIU implemented as a 3D LUT, so that should make for the most of the bulk in the file.
The guard rail ensures the chromaticities are preserved when outputting to a narrower gamut footprint.
1 Like
There’s also the fact that I used a different method to build the LUT this time: instead of manually translating all the steps into equivalent clf operations, I used ociobakelut to do the heavy lifting, plus matrix operations and power and log encodings to convert from/to ACES2065-1. That’s much easier to do but the price to pay is that the bulk of the ops are encoded in a big 3d LUT, which in principle causes some loss of precision (though I didn’t notice any visible degradation so far)
1 Like