no preview
DESCRIPTION
https://gofile.io/d/hIRzpc
ItBeginsAgain
Same release folder for everyone, look for Bastard_v20dpo.safetensors`.
This is a DPO (Direct Preference Optimization) trained SD15 model, taking the last premium release and applying 2000 steps of DPO training, which results in more correct generations across the board.
This is a preview model, and if successful I will conduct a 4x larger, 4x longer training run and see subsequent performance.
From initial tests, it appears at the moment that I should DPO train every version I release, as its just ... better haha.