It s hillarious that amd console owners think they will get this tech anyway.
Every argument I've seen against it is purely emotional, none of them are technical.
People complained about ray tracing when it first came out. Today it's the industry standard and nobody complains anymore (maybe console gamers and AMD GPU users).
And about "artistic intent", isn't this also messing with the "artistic intent?" Because this looks far away from the original intended art.
nah we already have plenty of games that do photorealistic graphics with real life actors. Our brains dont reject them because they are done well. Starfield and Zora AI humans look wrong for the same reason day time Soap Opera looks off. No one wants to play real life.That, but also I think the issue is that virtual humans are just not cut out for attempts at photorealism. As we're edging closer to that, it's creating this uncanny valley effect that audiences aren't comfortable with. DLSS5 could be as far away from slop as it could be, and I imagine a lot of people would still be repulsed.
I think chasing photorealism is fine in game environments as our minds are not trained to spot when a shadow doesn't cast just the right way. That's why rasterization tricks have gotten us so far. But our minds are trained (probably through evolution) to spot and warn when a person or creature doesn't look right.
I think this is why Rockstar are still leaning into their characters looking just stylized and 'videogame-y' enough so that they step past the uncanny valley.
I think nvidia engineers have kinda misread the market here. This is currently happening at every major tech firm so im not surprised that the execs and engineers at nvidia are also in their own little bubble.
Here is the thing, gamers largely praised games that went with a photorealistic art styles and character models. Hellblade 2, Matrix and Marvel 1943 were roundly praised by the same people showing an absolute disdain over this. I still think it's largely due to the AI faces completely missing the original lighting pass resembling Sora AI videos than the original game. But it's still a valid concern that apparently nvidia engineers and marketing guys were completely oblivious to.
The examples below showcase that people are enamored with photorealism in video games. This is possible on a 10 tflops card. We are going to 30-40 tflops cards next year. We can easily get to photorealism organically without taking any shortcuts. I think Nvidia just didnt realize that we had to get there ourselves.
You didn't address my point but I understand what you mean. I think it is important to dissect what people mean by graphics & art, they're already struggling to separate games with photorealistic art VS what they deem as "art", which is any game that doesn't emulate real-life.Nah, this is just bad graphics done by developers who clearly dont give a shit about character models looking good during gameplay. You can easily fix this with hero lighting. We dont need an AI pass to fix stuff like this.
This is what Grace's model looks like in model viewer. It looks nothing like what the Nvidia AI model's interpretation. Or rather hallucination.
I thought I did address your point. You said that they dont have control over Grace looking like that in the raster version. They do. They just chose not to. They arent the only ones, we have talked about it ad nauseum in this thread. TLOU1 Remake discussion in this very thread pretty much revolved around the characters looking like dogshit during gameplay.
She still looks a porcellain doll compared to the best a ps4 can offerNah, this is just bad graphics done by developers who clearly dont give a shit about character models looking good during gameplay. You can easily fix this with hero lighting. We dont need an AI pass to fix stuff like this.
This is what Grace's model looks like in model viewer. It looks nothing like what the Nvidia AI model's interpretation. Or rather hallucination.
ive always said that her character model in the game looks last gen. hell, she looks last gen compared to Leon.She still looks a porcellain doll compared to the best a ps4 can offer
You genuinely can't think that grace represent a realistic woman face in her current version.
You sincerely think devs don't have control? cmon man let's not go nuts with the conspiracy.
Hellblade 2 has some almost completely photorealistic moments.That, but also I think the issue is that virtual humans are just not cut out for attempts at photorealism. As we're edging closer to that, it's creating this uncanny valley effect that audiences aren't comfortable with. DLSS5 could be as far away from slop as it could be, and I imagine a lot of people would still be repulsed.
I think chasing photorealism is fine in game environments as our minds are not trained to spot when a shadow doesn't cast just the right way. That's why rasterization tricks have gotten us so far. But our minds are trained (probably through evolution) to spot and warn when a person or creature doesn't look right.
I think this is why Rockstar are still leaning into their characters looking just stylized and 'videogame-y' enough so that they step past the uncanny valley.
This is the future.Have you people completely lost it?
Do you think grace looks as good or better than these? It doesn't even look close. it looks 10x worse.
Its almost like everyone in this thread has had collective amnesia.
There are no free lunches, either you spend the 2nd 5090 equivalent power rendering the path tracing 2.0 environment below or you spend it on your beloved AI slop filter.
You can't have it both ways. People say, well its optional you can just turn it off. These people need to turn on their brains.
Its NOT optional. When DLSS was added it took up a significant portion of the die space that would have gone to traditional gpu pipeline.
Now it was worth while and a worthy cause from Day 1. But just turning it off didnt remove the tensor cores from your gpu and give you those removed gpu. The same way the 2nd 5090 equivalent tensor cores have now been reserved for this AI Slop.
So rather than spending the power of that tensor cores for the below. it will be spent on AI slop.
So turning it off just means ALL THAT die space on your GPU become useless because they have been reserved for AI slop.
They dont have control over the actual model. They have control over the intensity sliders which will allow devs to reduce the photorealism of the character models. We will see if devs can make it look non AI sloppy. But so far the results in virtually every game by every developer look very slop-ish.
It´s a model reimagining pictures plus some buffer values. It´s not exactly a straightforward process to finetune these models in any way. Maybe Devs get the possibility to do specific trainingssets based on what they want to see, but that would contradict the "general model" principle, who knows.
Adding a dog-snout and bunny ears to a person for a 30 second clip is so far away from looking at a "this probably looks like this"-AI video you`re interacting with for hours that I simply can`t express how nonsensical that comparison is.
The point itsn´t that she looks more realistic, if you think of that AI-slop fotostudio-look in that way, but that the AI completely changed the model while some retards are still claiming "it`s just the lighting". This isn`t enhancement based on the actual ground truth, it`s reimagination, that`s the point and the issue a lot of people have with this.
I was not talking about that type of filter, obviously.
SlimySnake
dude this was literally your first reaction to dlss5
the main issue, the very nature of this tech, will not change:
