>>105826842
I can see where he's coming from, though I'm kinda partial to the kl_optimal results. Shame they get so dark and grainy at higher CFG. Used 4 here.
Box: https://files.catbox.moe/h3ulib.png
>>105827088
Every 4 days, yeah. It does change, that's true, but the general gist should stay the same.
I'm holding off on a big "all schedulers/samplers" run until epoch 50, but I imagine someone else will do it at that point anyway. Just trying to see what I to use in the meantime.

Alright, I'll stop spamming my silly experiments now.