I don't think they're generated photos. They look like they might have been using topaz (or probably something worse/cheaper/online) to denoise and up-res-for-4k-TV really crappy digital photos from the aughts. The lower-res and noisier the original, the more detail the AI has to make up. Especially if they didn't have access to originals and were using photos taken from websites and reporting contemporary to the events, oof, there's only so much the tools can do.