For best outputs (both versions): Sampler: DPM++ 2M Karras. Steps: 20, cfg 5, CLIP skip 2. VAE: none required. Pos prompt: include the LoRa "add_detail" with a weight of 1 at the end of your prompt (included in my example runs for both versions). Neg prompt: "(worst quality, low quality:1.8), (illustration, 3d, 2d, painting, cartoons, sketch), bad teeth, open mount, (nude)". What changes from v9.0 to v9.1: input_image feature. Checkpoint Trained.