I asked DALL-E 3 for a picture of San Francisco with a cat terrorizing the city. It gave me a recognizable skyline and a cat that looked playfully destructive rather than frightening.
That was my reaction to one prompt, not a benchmark against every earlier image model. The result was good enough to make me think about using generated images to explore an idea before committing to it.
The useful question for my own work was how much editing the image needed afterward. Getting an amusing cat was easy. Getting the exact composition I wanted would take more than accepting the first result.