In my experience with SD, getting images that aren’t obviously “wrong” in some way takes multiple iterations with quite some time spent tuning prompts and parameters.