chrome musk
Like, I asked it to consider a design that would fit a certain market. And it included all of that information in the prompt it sent to DALL-E.
Which produced strange text on the image.
But generally speaking, I've long had the feeling that it doesn't know the exact capabilities of DALL-E, and what kinds of prompts are likely to work.
I don't know if the problem is still there, but in the past it would also put in descriptions assuming that DALL-E was maintaining context and knew what was being discussed in the chat... Which would usually prove to not be the case.
Maybe the DALL-E side prompt enhancer should also have a system for correcting against potential nonsense. But since image generation is a tool, it makes more sense to teach the LLM how to think about using it effectively.
solid cradle