Essentially confirms it basically is generative AI filtering over a 2D image based on questions given to an Nvidia rep. It basically treats the frame as a 2D image and then uses vector data which tells it which direction things are moving and then creates a "new" generative image from that. This is why the character from Starfield I believe clearly has hair added to an area of their head when there was no hair in the original model. It is ... basically an AI filter that reinterprets the original image and changes it.
Now it feels like Nvidia knew there was going to be blow back to this and tried really hard to sell this as "just a lighting change" (lol), when it's overall image manipulation because they likely are weary of terms like "AI slop" and potential blow back from artists working on these games.
But that does beg the question, if it's just image manipulation ... could they do something like photoreal image manipulation instead of trying to mimic existing video game graphics. And that opens up an entire pandora's box.
I also have doubts this will work great on sub 5090 hardware ... not because of the graphics complexity itself really but because it needs to time to essentially create the "new" AI frame and still be playable, which is probably why people mentioned it looked like the games were running at 30 fps. It has nothing to do with the game engine itself, it's just taking the image as a 2D input and reinterpreting that. But that also means if you're an artist on these games, essentially people are not seeing your work (your 3D model, textures, etc.), they're seeing basically a manipulated version of that image done by generative AI that is arbitrarily making artistic choices in effect.
Last edited by Soundwave - 2 days ago
