Overall improvement of stablediffusion in a single LoRA.
extraction + unet training + dpo merge
SDHK.safetensors