Actually, I’m also curious whether you’ve noticed whether there’s any pattern to how the effectiveness/accuracy of prompts develops during training? For example, does the accuracy of prompting evolve at a similar rate to image quality, or is it a refinement that happens at a later stage of training?