So, it occurred to me that there are captchas out there for improving things like OCR. What about creating something similar to recaptcha to improve GPT images. Maybe something that allows people to select images that don’t look right or something like that. Perhaps use keywords from the prompts and ask, “which of the following images matches” … [keywords from prompt], and another one where you know you have at least one right answer.
This might be a good deterrent for existing AI, while simultaneously helping you quickly get feedback about which images are not matching the prompts very well