I keep tweaking my prompts and even writing new ones, but this afternoon I had what I think is success. It really looks like these women are involved in a pie fight, and they look realistic.

I used ImageFX. All the ones I've used have strengths and weaknesses, and quirks. "Chest" works on ImageFX but certainly not on Bing. Krea just choked when I added "Asian." (and a few minutes later, accepted it.) If I don't specify "pretty" in ImageFX, the girls are sort of...plain and a little heavy. ImageFX is also the only one of the three I use for pies that regularly forgets the hand part of the gloves.

Anyway this prompt, )with variations in the dresses and colors, Yes, MessG) was:

A pretty young woman with wet hair wearing a soaked silver lamé strapless party dress and long gloves has pie all over her face and chest. She is picking up a pie to throw and looks determined.

The thing with the newer image models Like Flux, ImageFX etc is they really benefit from having descriptive prompts that specifically tells what you want.

I expanded on your prompt:

The image depicts a pretty young woman in a soaked, gleaming silver lamé strapless party dress and matching elbow-length gloves, completely overwhelmed by an extreme mess of gooey cream pie filling. Her wet hair is heavily saturated, thick strands plastered chaotically around her face and shoulders, coated with dense clumps of creamy filling. Her face is entirely buried beneath layers of whipped cream and gooey pie, obscuring her features and dripping heavily onto her neck, chest, and down the front of her soaked metallic dress. The once-glossy fabric is thoroughly drenched, smothered by dripping streaks and thick smears of whipped filling. Holding a heavily-loaded pie brimming with vibrant cream filling, her gloved hand is poised confidently to launch it, emphasizing her defiant, playful determination. The neutral backdrop accentuates the extreme cream-covered messiness and dramatic contrast between textures

Yikes! Lots of words. I don't think I've written anything that long since college.

After posting the first message, I tried the exact same prompt in Krea (Generating, not enhancing), with the results shown here. I thought the blue dress ones were best.

There WAS one change: I had to specify CREAM pie. ImageFX assumed that. Krea did not. But when I changed "pie" to "cream pie" both times, Krea wouldn't make the images at all!

Krea-generated images generally have to be lightened. These weren't too bad.

It's been a while since I've used Krea which used to use SDXL as it's main underlying model. I believe it now uses flux or has the option to.
The point is, SDXL uses an older language interpreter (Clip) for prompting whereas Flux, ImageFX etc use different language interpreters. Flux uses T5. ImageFX I'm not sure which but it behaves similarly.
In the old days, training of images were done on specific Key words. i.e. an image of a woman being pied may have been captioned with Woman, "polaroid, close up,40yr old, red dress, pie in face, Extreme Mess, cream"
These days, most models are trained on AI based captioning which is far more detailed and the model will learn the concepts. To get the most from the model you need to use the more verbose prompts.
That's not to say you won't get some results using simpler captions but they will be missing details and specifics, quality etc. You need to think about what image sets the model was trained on. Just asking for a Pie may give you what you want, but more likely it will not always infer that you want a large cream cake, gooey with vanilla cream. Instead you may get a pastry pie, no cream etc.

Commercial models will be training on millions of images and captions, They're better at deducing what you want but of course you will have filters to deal with. SDXL and Flux base models which Krea uses are trained on much smaller image sets so they have less context to learn.
The flux models and Loras I train look to add this missing detail back using additional images and captioning.