A beta tester roasted my prompt last March and now I write them half as long
I was cramming 400 word prompts into an image model and getting mush back, until a guy in a Discord said I was "fighting the model instead of steering it." Now I strip prompts to one subject and one style word, and my keeper rate went from maybe 1 in 10 to 6 or 7 out of 10. Has anyone else had a stranger's offhand comment totally flip how they work with a model?