[retrained: based on ptx0/pseudo-journey @ 4000 steps from stable-diffusion-2-1 baseline on 3300 images] + 9500 steps on 22,400 images, polynomial learning rate scheduler, batch size 4, 64 gradient accumulations, FROZEN text encoder, 8bit ADAM, ZERO PLW (no regularization data), followed by 550 steps with unfrozen text encoder and constant LR 1e-8

13000 steps: trained from ptx0/pseudo-journey, polynomial learning rate scheduler, batch size 3, text encoder, 8bit ADAM, ZERO PLW (no regularization data)