Support NeoGAF

[viveks86]

If context is being fed or not I think is secondary to me. If it doesn't match the creator's intent but the public prefers the result with AI is it really a problem?

In my opinion, yes. It is a problem. Because that takes the emphasis away from the art and the artist to just pure consumption and gratification. It gives the reinforcement signal to corporations that we will consume anything as long as it is gratifying. And that's a slippery slope. Why even have an experienced artist anymore? Get an intern to sketch some stuff up and get AI to render it beautifully for you. Generative AI can do the sketching part too. So you wouldn't even need that intern anymore. And so on. That slippery slope runs the risk of destroying the industry in the pursuit of efficiency and dollars. Human ingenuity loses value and we would have no options but to consume entirely machine generated content. What good is that for human progress and flourishing? People are free to do what they want, but I would actively discourage this path. That's not to dismiss AI altogether, but to provide a feedback loop on what directions are better than others.

Use AI to accelerate path tracing by understanding the 3d objects and light sources both on and off screen and the artistic intent from reference/ground truth. Not to conjure it out of a single frame where it has no clue what's occluded or off screen. They are literally selling its deficiency as a feature. That type of shallow implementation should be discouraged so that research energy goes in a direction that isn't harmful to all the progress made so far.

The fact that a polling thread is being created for an option and the majority votes in favor of it suggests people seeing this as the way all games will eventually be made. So I'm doing my part to say there are other ways that can still use this tech, but not in such a shallow manner that spits in the face of real art.

Would you say the same about a mod or would you consider that differently because there is another creator behind a mod?

No I wouldn't say the same. Because that's a mod. The original creator and their creation still exists. The way the current tech is being positioned will eventually eliminate the need for the original creator.

 

[Trilobit]

 

[Diddy X]

I'm satisfied with current-gen visuals: Resident Evil looks fantastic! The real issue is that physics and environmental interactivity have completely stalled. Too many game worlds feel like static dioramas where nothing moves or reacts.
Developers just don't prioritize it because flashy graphics sell games, while physics don't pop in screenshots. It's much easier to market "eye candy" than complex systems you actually have to play to appreciate.

Also, I don't get why so many players here obsess over photorealism. To me, realistic games are boring! When I want realism, I just go outside.

I'm also satisfied with current visuals, but as you said, the industry will keep pushing graphics forward and this seems like the most viable path right now. AI improving videogame visuals was bound to happen sooner or later.

 

[MCplayer]

only for sim racing games or racing games with a focus on photorealism

 

[A.Romero]

In my opinion, yes. It is a problem. Because that takes the emphasis away from the art and the artist to just pure consumption and gratification. It gives the reinforcement signal to corporations that we will consume anything as long as it is gratifying. And that's a slippery slope. Why even have an experienced artist anymore? Get an intern to sketch some stuff up and get AI to render it beautifully for you. Generative AI can do the sketching part too. So you wouldn't even need that intern anymore. And so on. That slippery slope runs the risk of destroying the industry in the pursuit of efficiency and dollars. Human ingenuity loses value and we would have no options but to consume entirely machine generated content. What good is that for human progress and flourishing? People are free to do what they want, but I would actively discourage this path. That's not to dismiss AI altogether, but to provide a feedback loop on what directions are better than others.

Use AI to accelerate path tracing by understanding the 3d objects and light sources both on and off screen and the artistic intent from reference/ground truth. Not to conjure it out of a single frame where it has no clue what's occluded or off screen. They are literally selling its deficiency as a feature. That type of shallow implementation should be discouraged so that research energy goes in a direction that isn't harmful to all the progress made so far.

The fact that a polling thread is being created for an option and the majority votes in favor of it suggests people seeing this as the way all games will eventually be made. So I'm doing my part to say there are other ways that can still use this tech, but not in such a shallow manner that spits in the face of real art.



No I wouldn't say the same. Because that's a mod. The original creator and their creation still exists. The way the current tech is being positioned will eventually eliminate the need for the original creator.

I understand what you are saying but I disagree.

I don't think work is valuable just because someone did it. I mean, we spend the day criticizing games here and it doesn't matter if it hurts the creator's feelings, hell some people seem to enjoy doing that. To me it sounds like lowering the entry threshold and allowing pretty much anyone to be a creator. Can you imagine how many ideas live in people's minds but they can't see the light of the day because it's so difficult to make them a reality? With tech like this people don't need to be an expert yet they can create things like one. I think this discussion goes beyond DLSS 5 because if I understand correctly, DLSS 5 is not creating anything from nothing, it's just infusing lighting and materials... If it was creating out of nothing without any consideration of the real scene it could paint a car on top of someone's face.

Does DLSS 5 automatically eliminates the original work? Because if not, then your argument for mods still applies: the original creator and the creation still exist.

Nothing I've seen out of DLSS 5 so far seems to suggest that the original creator would be eliminated. GenAI in general in it's current form just changes the way something is created but in the future it might eliminate the need for a creator. However, that doesn't mean that the creator won't exist anymore. To me it sounds like if when creating the synthetizer and being capable of simulating a cello's sound automatically wiped every cello player in the world when in reality just opened the options for creators.

 

[BackBackPunch]

My undertanding is that DLSS5 is an AI process that improves how lighting behaves. It does not changes the geometry of the models or the levels. And if it does changes some aesthetic it through the devs.

Some might look good some might look bad. It depends on the game and how its implemented. I can see racing games, and games that pursuing realism being the most beneficied.

And it is a optional feature that right now is running on two RTX5090. The tech is awesome, if you don't like it don't use it. The discussion around if it should be implemented, and how impact develompmet, artistic vision and so forth is another complete topic for itself.

Thats what the guy at Digital Foundry said, but there's more nuance to it.

The game engine renders a frame, sends the frame along with color and motion vector data to "anchor" the AI output so that it doesn't insert random hallucinations, and then it creates a new AI generated frame (which is what you see). The final output isn't simply the original 3D models with better lighting-- it's an AI generated frame based on the original frame, using contstraints. If you look closely at the Resident Evil example, her face has changed in shape. Fuller lips, differently shaped eyes, etc.

 

[phant0m]

Also, you guys know DLSS from the start was basically baby's first GenAI, right? Deep learning, neural networks, creating new pixels (or later frames) out of compute?

 

[Nintendo's Witch]

LMAO no

 

[GermanZepp]

Thats what the guy at Digital Foundry said, but there's more nuance to it.

The game engine renders a frame, sends the frame along with color and motion vector data to "anchor" the AI output so that it doesn't insert random hallucinations, and then it creates a new AI generated frame (which is what you see). The final output isn't simply the original 3D models with better lighting-- it's an AI generated frame based on the original frame, using contstraints. If you look closely at the Resident Evil example, her face has changed in shape. Fuller lips, differently shaped eyes, etc.

Understadable, there is a lot more to it in the technical aspect, lot of nunces. However, the resident evil girl model doesn't change.

Look it at for yourself. EDIT; if it has nicer red lips is because capcom wanted that way, or maybe the previuos in game light was very flat. I don't suscribe with the tone of the tweet tough, but i do like busting balls with other fellow gaffers,

 

[uncreativename]

Understadable, there is a lot more to it in the technical aspect, lot of nunces. However, the resident evil girl model doesn't change.

Look it at for yourself. EDIT; if it has nicer red lips is because capcom wanted that way

Yeah, it's not changing the geometry. Not sure why people think that.

But it is changing a lot of the colours. That image in particular, Grace's hair colour is totally different, and her eyebrows, for instance.

Someone upthread posted a pic and said that dlss was making Grace look much closer to the actual actress, which was a wild take, because when I look at that image I see exactly the opposite. The original is way closer.

 

[PaintTinJr]

Nvidia said this a demo and DLSS 5 will run on a single GPU, and most likely a 5090 at 4K. This will no doubt be make to work on something like the 5080 at lower res with VRAM being the cap. Plus 2nm TMSC is going to give 20-30% power reduction vs 3 and Blackwell is TSMC 4NP.

I'm sure the 6000 series and beyond are going to be massively and more performant in not only compute but also increase in tensor performance. Never mind what comes after. If we look at leaked PS6 and Xbox Helix specs we've already got the VRAM sorted. In a couple generations, its not far fetched that we could see this trickle down to consoles. Just by an educated guess..

For it to be viable to be mainstream like DLSS4.x it needs to run on a single RTX 4070 at 60fps as a worst case, and it will be miles away from that.

At lower resolution the realism and hyper realism will be lost and if it results in 30fps too on a single GPU costing north of £1500, the solution you are describing will have no viable market. when none of the use cases they've demoed are suited to user gameplay.

Personally I think this is pie-in-the-sky as the future - even for Nvidia's DLSS future - unless the training for an extra 12months can get it running on far more of the existing RTX discrete GPU install base.

 

[Topher]

 

[MacReady13]

I think it looks great! Fuck the haters and these weird dumb fuck "AI is crap" bullshit.

 

[jshackles]

Irrelevant. At this rate, I'll never be able to afford it anyway.

 

[Topher]

Irrelevant. At this rate, I'll never be able to afford it anyway.

Yeah.....just saw the report from DF that it took 2 5090s. Whether folks like it or not, it's completely subjective, but what isn't subjective is the thousands of dollars apparently needed. This whole thing has been a waste of time. But at least we got some funny memes out of it, I guess.

 

[bender]

 

[Topher]

Can't take credit for it. It was posted on twitter.

 

[bender]

Can't take credit for it. It was posted on twitter.

Old people shouldn't use social media.

 

[Topher]

Old people shouldn't use social media.

 

[Zacfoldor]

Bender is AI and the man in the suit is human so I take it you agree with him.

 

[KenaiPhoenix]

Nah it looks kind of weird.

And not only that, assuming that this works and is adopted, this is the death of style. Now everything will just look the same.

 

[bender]

Bender is AI and the man in the suit is human so I take it you agree with him.

Bender is love and the "man" in the suit is the moon. Had he been a human, I would have exterminated him.

 

[BloodFree]

Is that a problem though? You have two plates of really delicious food. One is made with proper ingredients that people like and the other is made with bugs ( LOL). Both look and taste good. Which one are you picking? Cause I know for sure I would still pick the proper one.

 

[Polaris]

"Bad" human art > "Good" A.I. "art".

 

[Topher]

Is that a problem though? You have two plates of really delicious food. One is made with proper ingredients that people like and the other is made with bugs ( LOL). Both look and taste good. Which one are you picking? Cause I know for sure I would still pick the proper one.

lol.....I'll just say I'd be pickier about the food I eat than the graphical magic being used in a video game.

 

[BattleScar]

Lol.
Like pigs at the trough.

 

[bender]

lol.....I'll just say I'd be pickier about the food I eat than the graphical magic being used in a video game.

I'd rather Topher eat a cockroach than play a video game that approximates authorial intent.

 

[BloodFree]

lol.....I'll just say I'd be pickier about the food I eat than the graphical magic being used in a video game.

Sure, I get that of course LOL, that is a wild example.

But that is still an emotional response and that's the point. You just don't like the bugs even if they're safe to eat and have been prepared properly. You don't care, it's still disgusting to you for whatever reason. Replace bugs with AI here and it's not too far off I think.

 

[MiguelItUp]

I personally don't care for it at all. Especially with the examples they've shown, it's got that look that practically every AI generated image has that immediately tells you it's AI. That weird glossy sheen look. In motion they all looked even worse to me. Especially when it was showing a before and after. It felt extremely uncanny valley.

I still think that Grace image is atrocious. Hell, in both of the images of Grace that were presented on DLSS 5 they don't look the same. You can see obvious differences between the faces when comparing them. Especially around the eyes and brow area. I thought the first one looked awful, the second one doesn't look AS bad. Leon, however, didn't look nearly as bad to me, he also didn't look nearly as filtered as Grace, especially in the face. His example looks more like what I would think and expect out of the next DLSS. Improvements on lighting, shadows, fidelity, etc. But not completely reconstructing his face.

Regardless of how I feel about this, I can just not use it, and it doesn't exist. Which is what I'll do, lol. I will say I really wish we'd put this much effort (hell, MORE) into improving performance and optimization of engines. It feels like it's only getting worse and worse. It's a shame that when a game is optimized and performs well that it feels like such a rarity these days.

 

[Topher]

Sure, I get that of course LOL, that is a wild example.

But that is still an emotional response and that's the point. You just don't like the bugs even if they're safe to eat and have been prepared properly. You don't care, it's still disgusting to you for whatever reason. Replace bugs with AI here and it's not too far off I think.

Not sure I agree with the analogy completely but I get the point. Either way, the meme wasn't meant to be taken seriously. Some folks genuinely don't like the way this stuff looks and that's completely fine.

 

[Lord Panda]

They should only allow DLSS 5 for Bethesda Creation engine games.

 

[GermanZepp]

Some people are using AI to make memes about AI.

 

[viveks86]

I understand what you are saying but I disagree.

I don't think work is valuable just because someone did it. I mean, we spend the day criticizing games here and it doesn't matter if it hurts the creator's feelings, hell some people seem to enjoy doing that. To me it sounds like lowering the entry threshold and allowing pretty much anyone to be a creator. Can you imagine how many ideas live in people's minds but they can't see the light of the day because it's so difficult to make them a reality? With tech like this people don't need to be an expert yet they can create things like one. I think this discussion goes beyond DLSS 5 because if I understand correctly, DLSS 5 is not creating anything from nothing, it's just infusing lighting and materials... If it was creating out of nothing without any consideration of the real scene it could paint a car on top of someone's face.

Does DLSS 5 automatically eliminates the original work? Because if not, then your argument for mods still applies: the original creator and the creation still exist.

Nothing I've seen out of DLSS 5 so far seems to suggest that the original creator would be eliminated. GenAI in general in its current form just changes the way something is created but in the future it might eliminate the need for a creator. However, that doesn't mean that the creator won't exist anymore. To me it sounds like if when creating the synthetizer and being capable of simulating a cello's sound automatically wiped every cello player in the world when in reality just opened the options for creators.

Let me start with a note of agreement. We both agree that AI based tools are the future, no matter how you slice it. We just disagree on the nuance, of what slices are good and what are harmful.

Firstly, I agree that the synthesizer did not wipe out every musician in the world, but it wiped out quite a lot. It was massively disruptive. But the Synth also opened up whole new job opportunities and paved the way for new sounds that are completely unique to the creator. So what new job opportunities do you think DLSS 5 will create? What new lighting? How would, say… Kojima's game look different from Miyazaki's game that would look different from Rockstar or Naughty Dog or Atlus or any studio with an art department worth their salt? How would the "look" be any different from a talentless hack, like if I made a game? How do I synthesize new looks that are uniquely me with a bunch of predefined sliders built into the SDK? Everyone can adjust intensity sliders and masks. They teach that in 8th grade these days. How do I control my creative vision for every 3d model, light placement and mood if Nvidia hasn't already built it into their model? How do I input or describe occluded or offscreen objects if it's only analyzing the current frame? The approach is devoid of any nuance and requires a complete rethink to be useful for practical application that sustains people's interest past the initial novelty.

What would actually get me excited is we get the synth version of an AI based graphics engine. Where I get to author every note and call it my signature. DLSS 5 isn't a synthesizer. It's a post process overlay you choose from a menu where you can't even make one of your own. Choose film noir versus photorealistic versus anime, adjust masks and intensity and you have to live with whatever you get.

That's my critique of this approach. It's not that AI is bad. But AI can certainly be used in bad ways and I believe this particular "fast food" approach is bad. And if they ever make a real AI based "graphics synthesizer", I'll absolutely applaud it. That will disrupt the industry too, but it will enable new forms of creativity and content instead of just stifling existing ones.

 

[photogaz]

I'm torn to be honest. On one side it looks amazing but on the other I'm concerned about the artistic vision for games. Some games aren't meant to look that real.

 

[DKehoe]

 

[Three]

That poll result just reveals how loud the vocal minority is…

More people are saying they don't like it. It's just that "No" and "I don't like it" are split into several options.

 

[LiquidMetal14]

More people are saying they don't like it. It's just that "No" and "I don't like it" are split into several options.

And it's cool to hate things that you don't understand. It's like endorphins and dopamine. Who needs games when you have this high of outrage at things