Version 9.0 was created using multiple resolutions during training, which seems to have strengthened the LoRA's ability to apply a style without significantly changing the composition of the images. Prompting still follows version 6.0, but prompt adherence when using the LoRA is improved.