Generating an Adobe Express illustration with a single AI prompt can be frustrating! Save time and give yourself more control by planning your image as a combination of several layers.
For example, the complex single prompt
A swordfight between a princess wearing trousers and a baker wielding a baguette; setting is a castle hallway.
gave this as the best result:

At least someone is wearing trousers and someone has a baguette, but the details are all wrong.
But I can set the princess, the baker, and the hallway in separate layers and combine them.
For the princess: A princess wearing trousers and brandishing a sword, side view; medieval Europe

For the baker: A baker holding a baguette like a sword, full length side view; medieval Europe

And for the castle hallway: A castle hallway

Then I removed the backgrounds behind the princess and baker, erased a few unwanted sections, flipped the baker image so he faced the other way, combined the layers, and adjusted layer sizes and lighting to get this result:

Still room for improvement, but I have all the key components I wanted showing in the image.