The prompt was "at a party in the rain, a cream pie splatters on the face of a woman with wet hair wearing a corset, stockings and long gloves, her necklace leaving an outline in the mess." It has worked before.

Take a look at this afternoon's results (before Bing quit producing anything for me). Do you see any corsets? Gloves? I don't even see legs much less stockings.

I guess I should be grateful they are women with pies. It could have been ugly old men with t-bone steaks.

It seems like the tried-and-true Bing formula no longer works. At one time, you'd start describing the scene, then the wam portion 'hidden' in the middle, followed by the background, camera settings, etc.

Now I find I can start with the wam portion and add the rest, but I have to repeat some of the wam details towards the end or they'll get ignored. Even then, it's hit or miss.

I've written this on another post but I think Bing is being looser with how it rewrites the prompt, leading to more variability between different creations.

One thing I've noticed in your prompt is that you have a lot in one sentence that might work better as two sentences.

Instead of


You could try


My hunch is that it's taking the clause "woman with wet hair..." and disregarding the other information in the sentence, perhaps because there is ambiguity about whether "woman" or "wet hair" is the subject.

Thanks. I tried your version with the shorter sentences, and got ... confetti. One of the pictures did have a corset, though. Then I got 6 or 7 Dogs in a row.

With my prompts I've been getting confetti sometimes as well. And with the same prompt sometimes not. Just like your confetti images - confetti instead of mess.

I mean, confetti does make a mess, just not the kind we're interested in. I think this is the prompt rewrite, and I think it's only a Bing issue rather than a wider Dalle-3 thing.