Not an artist or very familiar with AI generation but to avoid needing 6000+ images a format like this might be reasonable:...
Something says we should list key scenes that there are no images (And yet have never been) for at this point, even if not one for EVERY screen, though with the AI functionality...I can see it getting there more viably than ever before.
1. Starting scene as described by the game(Sonic reading book to Cream in her room with variants on the book depending choice if not too hard to create).
2. Slightly changed character positions/arousal states(Cream falling asleep or Sonic with bulge in clothes and wandering eyes).
3. For any sex scenes an establishing shot which can be modified to have different expressions or fluids as the scene progresses.
4. A climax image for Sonic and/or Cream that is either another modified starting shot or sometimes a new image depending on the scene(Cream asleep with a facial shot after the cartoons/movie in living room or her post-orgasm on her bed after using the carrot toy.)
5. Final image of reaction post-scene if required with a few reusable images based on the outcome and stats gained/lost during it.
Having a few images be repeats for poses and expressions that are frequently used and then do 2-5 custom scene images seems like it would be easier than trying to make each scene 100% unique.
Could have fun with some custom Bad End type images of Cream kicking Sonic in the chest/groin/face or a few of her screaming for help and running away. I think this version had a lot of the darker non-con stuff edited out but there may be a few of those scenes left as well if anyone wants to try their hand at them.