say i had so much footage, and I manually (tediously) went through it and separated all the objects from each other and said what they were doing. Then, I could take these words and then generate images using words. Then all you would need is symbolic ai and you could generate logical sequences of images, puzzled together out of different bodies. so you could get cleverbot, and get it to generate images for what it said, (if it spoke in descriptions rather than just chat replies) or maybe it plays sharades as it talks.
The best symbolic ai to date is watson, and im just wondering about perhaps giving it a dreaming ability? out of a puzzle of sprites.
Its a pretty easy idea to mention, but hard to implement, of course, but I dont see why it wouldnt work, then you could have a watson ai as a game mediator, and if you had enough footage, then the game visual could be quite varied.