Here's the training data I used. The pictures are of the Edwardian actress Maude Fealey, largely because I think she's really good looking.
These are the outputs along with my prompts, in the order I generated them:
Overall, I'm not as impressed as I was hoping I would be. The likeness is hit and miss, and hands are no better than they are in other AI image-creation systems.
In order of least-bad to most-bad, I'd say: newsreaders, office worker, Mucha, WoW and impressionism.
Maybe I'll wait awhile before paying them some money for another go.
Alternatively, I could just use a different email address. Hmm...