Mastodawn

If you're wondering what's going on with me art / research wise, I'm LORA training all the things.

Trying it at 512 for 1.5, 768 for 2.1..

It seems powerful but I have quality concerns..

I think I will likely train my 1.5 Kompromat model on the output from my Disco Diffusion version so that I can bring back that gloopy weirdness..

Also all my genart.. probably

Recursive training seems to be the thing, you can create really solid prompts by finetuning their output. Bring stuff from DD to SD, from 1.5 to 2.1