Wanted to see if I could get a decent lora out of a single image, seems like it somehow worked pretty ok. The hair sometimes gets a bit messed up, though.
Lucy is a npc, robot character who appears in the first Professor Layton game with a very minor role.
Professor Layton has a lot of characters where only a single standing CG exists, I might do some others soon.
Upscaled it then manually traced over and redrew the whole thing for cleaner quality using Gimp 2.2.8.
I've also attached the redraw which was used in training.
Description
FAQ
Comments (1)
This really does help understand how SDXL/Pony learns. In particular it suggests to me many of my otherwise functioning models have problems replicating footwear consistently not because it's comparatively lacking in training data, but because footwear tags are relatively limited.
Edit: Default strength is set at 0. I think you missed the ".8" part there. I've posted a pic seeing how it interacts with another LoRA, changing clothes, and different artstyle (it kinda sorta got her hair?)