This aligns with my experience using image generators —- I can get them to generate really weird unique combinations of things (ie “an octopus dancing with a cat”) but when asking for a relatively common image with 1-2 unique aspects they seem to just generate the common image.

The folks who were able to get it by having ChatGPT generate a really detailed prompt is kind of interesting. When I’ve tried writing my own detailed prompts, it seems like the limit on detail is relatively low before it starts going completely off the rails.