This is just the beginning of this tech and despite the hilarious results, I believe it could be huge for gaming when this will be possible in real time.
Do you remember LA Noire? It's reminds me of that except that instead of the faces, they do this for the entire game. I feel it would lose its lustre the more it is used.
This looks sick. It’s still janky and not at all possible in real-time for now, but it gives an interesting glimpse into the future. Letting the filter handle the image while allowing the game-logic to manage simulation consistency (something LLMs still aren’t great at) seems promising.
www.neogaf.com
Also, found some better looking ones:
Edit: this is still some ways off. Its too expensive on a individual consumer bases, to slow, and can only interpret flat images (can't see geometry)
But if it gets implemented with in the game-logic, it has some real potential. But that's at least 5 to 10 years off.