Trained on MOSS-Image-Base.
Use a LoRA weight around 0.6 - 0.9
xiaoyu-krea-2-base-epoch-10.safetensors