Sometimes — actually, quite often — the AI will ignore certain parts of the prompt, and you need to nudge it a bit to do your bidding.
The following prompt is also actually quite bad. I tried to explain what I wanted and I probably over-explained it. This happened to me a few times since, too.
Me: Create a photo that includes a Surface Pro 9, in a horizontal position with its screen looking upwards, and add four squirrels figure-skating on the surface of the screen (as if it were an ice rink).
At this point, DALL-E ceased to give long explanations before the images. Instead, it went straight to the point:
See? The squirrels are in the screen, not on the screen. DALL-E omitted that bit. Also, there are five, not four, squirrels on the first image.
I had to give feedback and ask for a new batch:
Me: Re-generate the previous photo with the squirrels figure-skating on top of the screen, not in the screen.
DALL-E more or less got it right in only one image. The rest of the images reflected various degrees of hilariousness. The AI, like its predecessor, still struggled with the numbers. Later, I also learned that it does not play well with negative clauses (see “not in the screen” above).
thirdtower.ai 