User LoRA
Trained from Necomi's work, based on PonyXL v6V02, re-did dataset, trained with smaller dim