What's fun--if you're into some fairly harmless time-wasting--is to use the same keywords over and over while selecting the different styles. I had one set of keywords involving "gentlemen" and every style, including the Japanese one, delivered a vague sense of figures wearing black with white collars and cuffs.

So you can sort of guess at how the AI "thinks." I imagine stuff like this will continue to get better and better.