trained with 1500 pixiv img for style from v2
dylora dim64 alpha16merge with dim32
kohakuV3_rev1.safetensors