Those using Stable Diffusion use a huge number of words to produce their images, and it often looks overworked. The model is probably very different for that one.

Good results from Dall-e3 requires a "simple vagueness", this one has around 15 "directions", specify more and it turns bad or the filter blocks it (no idea why, since it doesn't show what was blocked).