Cover-Prompt Generation, Not Cover Images

The three question rounds are fixed: style and title first, title at four to eight characters, one to three candidates offered; then reference images, the first always a face and the rest product material; then four visual details chosen by number. One prerequisite has to hold: the image model must accept multiple reference images, or facial consistency is simply unachievable. The author also bans holograms, data streams and particle effects outright, on the grounds that they cheapen the picture.