There are whole courses and videos folks can take or look at to produce precise prompts to get the ai to act in a manner that you want.
I use “Text to Art” ai, and have to get pretty precise, with a lot of trial and error, to even get c,one to what I want as an outcome. I did some art prompts to create a .ikeness of Trump, and it kept creating results with him scowling or looking apathetic, so I had to refine the promt several times till it got what I had in mind, ie Trump smiling nicely in front of a crowd. Just putting in “smiling” didn’t work as the “smiles” were even kinda angry looking. Ended up with “smiling nicely while appearing happy” or somehting like that before it got good results
For an overview of one approach see https://learnprompting.org/docs/basics/intro
Nice story and reminiscent of my own experience. I burned through my free credits pretty quick and never did get exactly what I wanted, but close. It was an outdoor scene and the main problem is that the AI would insert random extraneous elements in the scene that ruined it. Like a big sort-of-a-cactus in the middle of an intersection. So then I was going to use another AI product to do a “magic erase” of the cactus but never got around to it.
And regarding your “smiling nicely while appearing happy” prompt innovation, I have done similar reiterations with ChatGPT. Kind of like you have to do with some people.