This was trained on three separate sets of training epoch sample images where the sample prompts were randomly determined by the trainer.