Finally, I have finished all the illustrations to "Talked Themselves Into It".
It will take me a month or so to sort them out and upload, but the most tricky (and interesting) part is done.
While I'm at it, an observation. Regarding the characters' appearance in AI-generated art, "regression to the mean" is a more serious problem than I thought. It often goes unnoticed, but I found out accidentally that Priscilla from this story (before her "blondification") often looks almost like a clone of Doris from "Movie Mania". Which was not an intended effect at all, obviously. Comparing their portraits side-by-side, I saw a striking resemblance, and got into it a bit more. As it turned out, characters had very similar text anchors. Both were described more or less as a "beautiful woman in her late 20s or early 30s, long straight dark brown hair, sophisticated features, high defined cheekbones, sharp refined jawline, almond-shaped eyes, full lips, statuesque hourglass figure", with some minor variations. So it's no wonder that AI starts depicting a generalized cookie-cutter "beautiful brunette" after a while. Nothing to be done now, but I'll have to take this into account from now on and dilute those general, standard descriptions with some more specific features (resting facial expression, ethnic anchor, asymmetries, celebrity look blending etc.)
