This is a difficult novel to describe in a few hundred words, which is why I resorted to the analog of a movie trailer. But there is a difference for novels. A movie trailer is literally a collection of scenes from the actual movie. The movie and the trailer arrive at the same time. A novel precedes the movie (if it gets one at all) by a few years, so the reader has a grace period in which to imagine the scenes and characters in their own mind. I didn’t want to preclude this reader prerogative. So I had planned to make the main characters in the iMovie trailer essentially anonymous, by always rendering them from the back, or ¾ profile, or via hazy reflections in mirrors or windows, so that you would know by the hair and the clothing which character was which, but you could still imagine the faces in your own way.

When I discussed this strategy with ChatGPT-4o at the outset, it thought this was a wonderful strategy. And it offered to help me find clever ways to obscure the main characters in scenes. But as I soon learned, once the project began, ChatGPT-4o is not the best advisor concerning what its companion image generator DALL·E is capable of. It is very good at explaining why the renderings fail – after the fact – but not so good at foreseeing that they will fail. So you learn this as you go.

One very strong bias that DALL·E (and indeed any current-generation AI renderer) has is to place main characters in the foreground of a scene, facing the camera – even if you explicitly tell it not to. This is because the AI has learned photo-realistic rendering from actual photos, where this kind of scene composition is over-represented. So Ilya and Katerinya had to have faces for this project to work at all.