retrained on anime model, with wd 1.4 captioning instead of blip. Works better with counterfeit.
Trained on 500+ loading screen pics, 1024x512
