>>105825856
Here's some CFG tests with euler/euler a and kl optimal.
Seems to be really sensitive to CFG, which is why the initial comparison is kinda fried (I usually use 4)
Catbox: https://files.catbox.moe/3m3qlx.png
>>105826225
Euler+Beta is what I mostly use, but I'm doing these tests to see if there's something that'd be better.
>>105826456
He has the default workflow in his HF which works fine, but I doubt it's the best the model can do.