I've messed around with this technology, and I find that 99% of the time you get uncanny valley looking images, and quite often anatomical anomalies. I'm sure if you have eons of free time to precisely hone in on which prompts produce which results, and give it lots of trial and error, you could make something decent. But the way I see it, the tech isn't quite there yet to be worth your time.
How those images in Null's post got made is beyond me. Using the instructions he provided, all I got was hot garbage. I couldn't even get more than one character to populate in an image, its like it was built solely to make portraits of anime girls.