LoRA is trained on vector portraits of different people, the recommended value is 0.8
vectorportraits_sdxl_lora.safetensors