This is a semi-failed LoRa, very experimental. It was trained on ~100 assets directly extracted from the official digital artbook, for 3000 steps.
Gen recommandation : er_sde with beta57 scheduler, 30 to 50 steps, CFG 5.5.
Don't expect too much from it.
Description
FAQ
Comments (3)
Very cool idea for a lora.
This is a good model, it gets the style very well, particularly if you guide it with styling tags. But it has a massive concept bleedover and tends to add bunch of stuff you didn't ask to everything it does.
Awesome style!