This is trained on nearly all of the datasets I've created, mostly as a test just to see the outcome of such a thing, however the results are quite nice, though the style can shift noticeably depending on what you're prompting.

Keywords

j3as4 - standard activation keyword

photograph, realistic, realism - helps trigger the one dataset based off of a handful of movie stills

detailed (and or background) - triggers two datasets that are far more detailed and semi-realism

I'm certain there's more, but the rest will be discovered with trial and error, these are just the most noticeable that come to mind.