This seems like a good effort. SDXL is however trained on a large amount of data with unknown, likely copyright infringing material.
Using a fine-tune or LoRA based on a consenting artists works doesn't remove the infringement of the others.
Consider the case where a licensed artist has a similar style to another with a much larger body of work in the training data. The model's output is going to be heavily informed by the latter artist. Essentially letting the former profit off the latter's work.
This is of course all assuming that AI output is copyright infringing.
Consider the case where a licensed artist has a similar style to another with a much larger body of work in the training data. The model's output is going to be heavily informed by the latter artist. Essentially letting the former profit off the latter's work.
This is of course all assuming that AI output is copyright infringing.