Eagle367 said:
Soundwave said:
I think what this will lead to is eventually a work flow where game's graphics are actually very basic, like maybe even PS2 or PS3 quality (if even that?) ... and then the bulk of processing spend is going to be to get the generative AI to quickly manipulate the image up to look like whatever photo/video reference it has. What DLSS5 is showing is it's effectively possible to do this in real time.
I have no idea where a company like Nintendo will sit on this in the long term, but theoretically it's possible to imagine a setup where it terms of raw compute, a Switch 3 for example is not really much more powerful than a Switch 2 on the compute side, maybe even less so, but all the "new" aspect of the hardware is a ton of added Tensor cores that handle the "image manipulation" (AI filter basically). Basically the silicon will be just a ton of Tensor cores dedicated to making the AI filter super fast and be able to extrapolate from some photo reference data it's been fed (trained on), the actual "real rendering" side of the chip is likely to be gimped because it won't be needed.
Because really why even bother with heavy native rendering when you can just focus the hardware towards basically becoming a AI filter machine. AI image manipulation doesn't care if you ask it to make something photoreal versus hyper cartoony versus hyper detailed or not detailed really. It's just manipulating pixels on a 2D image and then outputting it. Looks like it basically just needs to know from vector data which way things in the scene are moving and it can create a new image that "looks better" just from that.
But what that will do is basically ice artists out of the gaming process. Why pay an artist when the final frame is generated by an AI. It will eventually devalue graphics because who cares about graphics when its just an AI algo spitting out a 2D image filter basically, does anyone get really excited when they see more advanced AI filters that are more photorealistic going forward? I doubt it, it was neat maybe the first few times you saw it, now it's just a "whatever" or even annoying. |
That sounds bad to me because how can you control your art design and direction? Are you just gonna let an AI bot determine it for you? How will you ensure consistency? |
The AI basically decides based on whatever prompt criteria you've likely outlined (ie: "make this look moar real"). I suppose you can make it more specific by saying like "here's image data of an actor/actress, make the main model look more like that". That type of thing. But yes, essentially you are ceding control over your final image to an AI algo to essentially "decide" what the image should look like.
