As dumb as it sounds, it helps to write the descriptions as like a shopping list, so for example, what I used to get this was:

"A soldier in a purple trench coat and a white gasmask with purple visor shaking his fist angrily at a brass robot with a smooth face as it steals a jug of sweet tea and runs away."

It didn't quite turn out how I wanted it too, but you casually socking me across the jaw whilst holding a jar of sweet tea is pretty funny too.

Quiet_Boi's one actually gave me the most trouble, as I had to list every single possible detail to make it look right, so it ended up being this:

"A nervous, young Latino man dressed in a white t-shirt being hugged by a grinning, beautiful, red-headed female mercenary dressed in brass bulletproof armour."

I haven't attempted to get it to write text yet, so I'm not sure how well it'd work.

It really struggles with text from what I've found and it tends to be confused a lot of requests, but what I've seen could be used as a general template to make something better. More along the lines of themes and angles.

Input: "a brass robot with a smooth face and WW1 british helmet being confounded by Hieronymus Bosch's The Garden of Earthly Delights"

Result: