Retrained sock entrapment

Finally got OneTrainer working on my home computer and can finally tweak and test different promptings, training parameters, etc. Hopefully this means home stretch for the updated version of the main LoRA.

All of these are straight txt2img. No touchups. I think many would benefit with some inpainting on the tinies, which get a little blurred out due to the low resolution.

Observation - I think some of that style-shifting that I had blamed on overtraining is just what happens when you put "loli" in the negative prompt in most Pony-derived checkpoints sometimes. Even with source_anime, it tends to launch it right out of anime style and into a western cartoon style, even in an anime-focused checkpoint like autism.