• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

DLSS 5 - Generative AI

kac77 said:
People know that it might be optional now but there's a good chance it won't be in the future. You forget Microsoft bought a whole lot of game development studios. You think they will allow it to be optional? Not likely. Microsoft in particular needs to recoup its Open AI investments and this is a good way to do it.
Click to expand...
I don't understand this. How exactly will DLSS 5 from nvidia help Microsoft recoup their AI investment? It isn't like game development studios now just can save money by creating featureless dummies and have DLSS 5 fill in all the details, it needs the details to be present to work in the first place.
 
Eivind68 said:
I don't understand this. How exactly will DLSS 5 from nvidia help Microsoft recoup their AI investment? It isn't like game development studios now just can save money by creating featureless dummies and have DLSS 5 fill in all the details, it needs the details to be present to work in the first place.
Click to expand...
That's exactly where we're headed. nVidia ACE exists for a reason. Paired with their other technologies what you're describing is quite doable if it's not already happened in new development.
 
I think my main complaint is the AI filter is being baked into Deep Learning Super Sampling. I was under the impression DLSS was an upscaler. That demo didn’t demo upscaling. It’s a new feature and should be named and marketed accordingly.
 
I don't get the push back. Graphics have been stagnant for over 10 years. We keep throwing more and more TFLOPS at it for minor upgrades.
Nvidia creates an algorithm that makes meh materials realistic and implements realistic lighting and everyone is clutching their pearls. Neural rendering may end up as impactful as shaders and bump mapping were.
I honestly don't care what it does or how it does it. I want the game to look better.

I suspect the push back is due to AI causing memory and gpu prices to rise. I didn't know we were huge on respecting the art direction from AAA studios now days.
 
Also it appears that the developers (the actual artists) saw it the same time that we did. Article here.

1773846236205.png
 
Eivind68 said:
First of all I'm sure you can decide for yourself if you want to use it, and I think it looks like a notable improvement in many cases.
Click to expand...
more importantly maybe, only games that want to use it (i am sure there will be mods to force on game that don't), but some people act as if a mutli-billions game studio are being forced here.

Eivind68 said:
so why should they complain about their "artistic vision" being compromised after the customer has paid for it?
Click to expand...
why not ? here too, it is the game studio 100% in fault if that happen, it is the dev artistic vision (the game dev control if dlss 5 is used at all, how and how much)

Eivind68 said:
Do car designers cry if you change the wheels or repaint your own car, thus destroying the design they have worked so hard to accomplish?
Click to expand...
I am sure they do yes, specially if it happen to all cars being sold and not just some rare modders, Ferrari will not sell you an next one if you do this in a way they do not like, you end up on their blacklist.

Some butcher will not sales you again aged steak if they hear you having put ketchup on it, chefs can get angry if you put X on their meal they made before eating it and i am sure many game artists hate some mods that become popular, that human nature.

kac77 said:
In this image we see AI struggle to maintain the contours of the ball.
Click to expand...
Without seeing how motion blur of the dlss off look like, seem a bit hard to say.

kac77 said:
e points out what most of us can see, it's more than just lighting that's changing.
Click to expand...
Of course, that the very claim made by nvidia, that it is a neural generative shaders ?
 
Last edited:
Ranulfo said:
Hmm, given Todd Howards statement/video about the tech I'm guessing Bethesda devs knew. Maybe just Todd and the execs.
Click to expand...
Pusblisher all knew, higher up has well, but this seem to have been something they achieved to keep secret for an annoucement with I think zero leak ? in a very rich moneytised leaky tech world, I would believe many low level employee didn't know and that the demo is the worse version we will ever see instead of a particularly good one, has they did not used team of artist sematic-masking-intensity input on it.
 
Last edited:
equinox654 said:
I want the game to look better.
Click to expand...
Don't you think that before/after of the blonde in I think Resident Evil, the DLSS makes her look older/sick/tired, just from the lighting change? I agree there's some good detail enhancement, but my overall impression of that specific picture is it's NOT an improvement, unless it's an impression the developer actually wanted but couldn't do originally.
 
One big issue with the photo realistic being the first neural aim, it will work well with non/low uncanny valley items (cars, roads, ), could take a while like movie CGI to work well with face and every human brain are super focused on detecting minuati-off details.

Can see many game simply choose to not apply much of it on the characthers, but if the world get really photo realistic, now the humans that are not start to look strange to it.

A model that let receive as an input picture at least (or even the actual 3d model) from z brush, before the game made low poly version would be interesting to see (if not trained on them as a partial over it sub model).
 
1_rick said:
Don't you think that before/after of the blonde in I think Resident Evil, the DLSS makes her look older/sick/tired, just from the lighting change? I agree there's some good detail enhancement, but my overall impression of that specific picture is it's NOT an improvement, unless it's an impression the developer actually wanted but couldn't do originally.
Click to expand...
To me it looks like she went from being made of rubber to closer to photo realistic. I would prefer the DLSS 5 image.
The part that shows her actually in the shit would make sense for her to look older and tired. Maybe they only made one profile.
What about Leon? He just looks better lit with better skin detail.

I also don't see it as the final product. Who knows how the final version will look or how it will iterate. I also don't know how much time and effort capcom put into the presentation.
It looks fairly promising to me.
 
kac77 said:
Also it appears that the developers (the actual artists) saw it the same time that we did. Article here.
Click to expand...
I'm not familiar with the reputation of that site but it wouldn't surprise me if various artists weren't made aware and it was a higher up decision.

That said, is this feature something that can be added by merely swapping some DLSS DLL like the upscaling component could be in the past? Since Nvidia's messaging suggests it's something a game has to support independently, in which case it would need a game's own devs to implement support for it. The article specifically says 'devs' in relation to Requiem, which makes it sound either this can be forced into games that don't support it or those devs just weren't involved in adding support for it.
 
  • Like
Reactions: erek
like this
Okatis said:
I'm not familiar with the reputation of that site but it wouldn't surprise me if various artists weren't made aware and it was a higher up decision.

That said, is this feature something that can be added by merely swapping some DLSS DLL like the upscaling component could be in the past? Since Nvidia's messaging suggests it's something a game has to support independently, in which case it would need a game's own devs to implement support for it. The article specifically says 'devs' in relation to Requiem, which makes it sound either this can be forced into games that don't support it or those devs just weren't involved in adding support for it.
Click to expand...

You will be able to plop it in any game just like with other versions of DLSS. But it won't work as well without the developers wiring it up with the data it needs and tweaking things. Just like with other versions of DLSS.
 
equinox654 said:
To me it looks like she went from being made of rubber to closer to photo realistic. I would prefer the DLSS 5 image.
Click to expand...
I can understand that--I even said there was some good detail enhancement. I'm just saying if I were a real woman upon whose face that character was modeled, I'd be annoyed that the lighting made me look old. It's almost all what I would call unfortunate lighting angles causing it, too, which is stuff competent real world photogrphers know about: It's the shading on the folds from the nose to the corners of the mouth (I forget what those are called), and the shading on her cheeks & neck.
 
Okatis said:
That said, is this feature something that can be added by merely swapping some DLSS DLL like the upscaling component could be in the past?
Click to expand...
My quess it will be and the optiscaler of the world could find a way at some point, but it will be an all by default case with what it can do from dept field, motion vector and the render without added extra semantic (well maybe they will let the gamers add those), maybe it will be good enough to detect silk is silk, hair, furh, etc.. without having it added manually but a certain performance cost.

Could make the game run quite slow, because we can imagine that when dlss 5.0 is on, game shut down expensive part of complex shaders (like the type that make skins look like skins, water like water, foliage, etc...) letting dlss 5 do the work, now it will be ran twice possibly the complex skin shaders + dlss 5 skins over it.

Ideally you want to have semantic label there with your asset, a bit like https://docs.omniverse.nvidia.com/extensions/latest/ext_replicator/semantics_schema_editor.html, to define material class yourself and save where you can on pre-dlss compute.
 
equinox654 said:
I don't get the push back. Graphics have been stagnant for over 10 years. We keep throwing more and more TFLOPS at it for minor upgrades.
Nvidia creates an algorithm that makes meh materials realistic and implements realistic lighting and everyone is clutching their pearls. Neural rendering may end up as impactful as shaders and bump mapping were.
I honestly don't care what it does or how it does it. I want the game to look better.

I suspect the push back is due to AI causing memory and gpu prices to rise. I didn't know we were huge on respecting the art direction from AAA studios now days.
Click to expand...
You can think it looks better but I'll disagree. The Resident Evil shot makes it perfectly clear why it looks like shit. It's a dark, rainy day and she's standing between two tall buildings with the only light sources being from straight above or through the small area of the alleyway behind her. The moment DLSS5 is turned on it looks like she's standing in a studio lit up by floodlights all around her and the background is a CGI green screen. The lighting and the way she looks doesn't match the environment and looks like shit. She's not supposed to be lit up like she's on a runway at a fashion show. Not every detail of her face is supposed to stand out. Her face is supposed to look dark and shadowed because that's exactly what you would expect in that environment under those conditions and DLSS5 destroys that. It turns her into uncanny valley compared to the scene she's in and destroys the mood of the scene.
 
Just wait till DLSS 13 is released...

Anyways, it'll be interesting to see DLSS 5 in actual gameplay. Getting on hate or hype trains without experiencing something firsthand is dumb af.
 
Eivind68 said:
I don't understand this. How exactly will DLSS 5 from nvidia help Microsoft recoup their AI investment? It isn't like game development studios now just can save money by creating featureless dummies and have DLSS 5 fill in all the details, it needs the details to be present to work in the first place.
Click to expand...
You think this is the final form of this tech?
I think it should be pretty obvious where this is heading. The publishers (the money behind video games) of course are behind this. They are hoping gen 2 or 3 of this tech allows them to prompt NPC models, Prompt lighting. This is the path the replaces 80% of the studios staff.

This is also potentially the path that leads to games that won't run on that potato you bought last year to play video games on. So why not just stream the assets... or stream the entire thing from our service. Sure you have not been a fan of streaming before. But what if the streamed version was the only way to play the ultra quality mode? (and maybe the only way to play the game at all)

This might not be streaming yet. I would suggest however if Nvidia are demoing tech intended to play old games 6 months out from launch apparently and requiring 2 flag ship consumer GPUs ($6-8k worth of compute) to pull off a demo that looks like a reshade filter. It feels to me like they aren't all that serious about having this tech work on consumer GPUs locally. Though with the rumblings of Sony preparing Asset streaming for the next playstation. I think a picture is starting to clear up. imo If games are streaming assets anyway.... why not start off loading this as well. The next iteration of streaming seems to be getting ready to launch, they are going to sell it to use on two fronts. First saving space on downloads... no more 100gb game downloads, how about 15gb just the core, all assets will stream... after being beutified.

Or at least I'm sure the game publishers who are giddy about this tech are very much thinking that is where this is heading. This results in them making more money. Piracy being no more, monthly residuals, it even increases the number of people able to play "ultra" settings to almost everyone. Little downside for game publishers.
 
1773863189054.png


Pretty gross stuff. If this is the future of gaming, I'm out. I have other hobbies I can go back to tbh. I'm not playing a game that requires this tech. As long as it's optional I don't care, but games will require it, and I'm not interested.
 
ChadD said:
You think this is the final form of this tech?
I think it should be pretty obvious where this is heading. The publishers (the money behind video games) of course are behind this. They are hoping gen 2 or 3 of this tech allows them to prompt NPC models, Prompt lighting. This is the path the replaces 80% of the studios staff.
Click to expand...
There is no prompting involved here, this is just a V2V generative AI like any other. The only things dev can control is how far they dial up the slop. What this might allow the devs to do later is to basically make a game of placeholder quality assets that the AI will try to pass off as faux realistic. It's the videogame version of death of the author as the game developers artists and designers have no precise control over the end result. Which will make every game look the same that uses the tech.
ChadD said:
This is also potentially the path that leads to games that won't run on that potato you bought last year to play video games on. So why not just stream the assets... or stream the entire thing from our service. Sure you have not been a fan of streaming before. But what if the streamed version was the only way to play the ultra quality mode? (and maybe the only way to play the game at all)
Click to expand...
If you're only streaming assets it still has to run on your "potato"
ChadD said:
This might not be streaming yet. I would suggest however if Nvidia are demoing tech intended to play old games 6 months out from launch apparently and requiring 2 flag ship consumer GPUs ($6-8k worth of compute) to pull off a demo that looks like a reshade filter. It feels to me like they aren't all that serious about having this tech work on consumer GPUs locally. Though with the rumblings of Sony preparing Asset streaming for the next playstation. I think a picture is starting to clear up. imo If games are streaming assets anyway.... why not start off loading this as well. The next iteration of streaming seems to be getting ready to launch, they are going to sell it to use on two fronts. First saving space on downloads... no more 100gb game downloads, how about 15gb just the core, all assets will stream... after being beutified.
Click to expand...
Game streaming without significant lag while maintaining high image quality is magic. You seem to be misunderstanding the concept, DLSS5 doesn't deal with assets, it is an AI filter enhancing the rendered image. If it dealt with assets then it didn't need to work at run time, assets could be enhanced in advance.
ChadD said:
Or at least I'm sure the game publishers who are giddy about this tech are very much thinking that is where this is heading. This results in them making more money. Piracy being no more, monthly residuals, it even increases the number of people able to play "ultra" settings to almost everyone. Little downside for game publishers.
Click to expand...
Game publishers being giddy about it should ring alarm bells if anything.
 
M76 said:
There is no prompting involved here,
Click to expand...
this will be coming fast right now the semantic you add is only stuff like skin, furh, hair, satin, plastic, foliage, water (just 256 different right now).... but soon you will be able to add sentence and it will become multi modal and not far after what you will need to give from a gamer engine point of view could become really basic has well (will be interesting to see what it look like right now), not only control via text, but imagie, 3d model, sound and others.

M76 said:
It's the videogame version of death of the author as the game developers artists and designers have no precise control over the end result.
Click to expand...
game dev will make their own soon enough per games will be my guess (at least added to dlss), small indie will use pre-made (unreal, nvidia, amd, sony, etc.. one) but big one will make their own per franchise. Question will be either from scratch or extra shader above the DLSS one just for some elements they want.
 
DLSS is still lazy slop and creatively bankrupt. Nvidia is just trying to maintain a stranglehold over everything they can, and lazy developers are happy to ride shotgun to the bank. Although, hopefully it's off a cliff instead...

Screenshot_20260318-162123.png





Screenshot_20260318-162324.png
 
LukeTbk said:
this will be coming fast right now the semantic you add is only stuff like skin, furh, hair, satin, plastic, foliage, water (just 256 different right now)....
Click to expand...
If they do offer a way to give hints to the model like this devs would have to commit more to Nvidia proprietary tech, since that's not just drag and drop. Granted, there was that figure that the vast majority of the market is Nvidia anyway but does make me think of PhysX.

MikeTrike said:
snip
Click to expand...
KojiPro were early to writing a from-scratch PBR-based engine so they're very experienced but yeah they really know how to make things look good (which is also very dependent on art directed lighting, much like IRL).
 
mullet said:
All the podcaster did was talk about feelings and how the care that the artist put into that image was written over by Ai, mean horrible abusing Ai.

This crying about Ai stealing art from pure artist and overwriting some poor working artist trying to scratch out a living. HAS TO END!

What ever DLSSSSS-5+ extra did made the game graphics look tons better. Does Ai have some flaws in graphics like the soccer ball? YES but in 3 years you will not be able to tell at ALL!

Ai graphics is going to take over and there isn't anything anyone can do about it, some of the coolest graphics I have ever seen are from AI.
Click to expand...
You might like nVidia and that's fine, but you should be educated about what's happening here and WHY the images aren't good. You think it looks great but had you watched the video you would have known that a lot of lighting is stripped out and it's changed the feeling of the entire scene. In that scene the DLSS5 version has her looking like she's going to a photoshoot when really she's going to see where her mother was killed, which the AI filter completely stripped out. That's why it's bad. Same thing with Leon but I won't give that scene away in case you haven't played it.

Finally, photo realism has been done before without this technology. Just look at Death Stranding or any Hideo Kojima game. He's been doing it for years without the need for this tech. Does it have it's place? Maybe but in terms of for gamers in general? No. This tech is trying to replace developers and that's about all its trying to do. You want AI to take over development? You better be careful for what you wish for.
 
Also just to contextualize this progress, for those who may have missed it, years ago Intel demoed footage of GTA V with style transfer to look like dashcam footage. There were plenty of prior style transfer projects but this was temporally and stylistically stable. This got millions of views and people interested in the potential for a game that could ship with some baseline quality of graphics and be transformed on-the-fly.

So when I see things like pavement being smoothed it seems in line with such transformations that are aiming for some prebaked idea of what photorealistic means.


View: https://www.youtube.com/watch?v=P1IcaBn3ej0
 
M76 said:
There is no prompting involved here, this is just a V2V generative AI like any other. The only things dev can control is how far they dial up the slop. What this might allow the devs to do later is to basically make a game of placeholder quality assets that the AI will try to pass off as faux realistic. It's the videogame version of death of the author as the game developers artists and designers have no precise control over the end result. Which will make every game look the same that uses the tech.
Click to expand...
Not this iteration no. Though there is no reason the two could not be married. Though you are probably correct there may be more of that in dev tools to come but maybe not the engine. (Though I don't think NV adding a drop down menu of user select able "styles" is out of the question in a future version either) To me this seems like its basically a more compute intense Reshade filter. Having various lighting or warmth options I'm sure will end up baked in at some point.

M76 said:
If you're only streaming assets it still has to run on your "potato"
Click to expand...
No the DLSS5 stuff doesn't have to run on your potato that would be the point. If the game has already been setup to not keep any texture assets local anyway... why not process them before sending them.
https://tech4gamers.com/sony-patents-shrink-size-games/
This is a sony patent and maybe its just a patent on someones brain fart. Or this is somehow part of what Sony/AMD is cooking in this same line dlss5 thinking.

M76 said:
Game streaming without significant lag while maintaining high image quality is magic. You seem to be misunderstanding the concept, DLSS5 doesn't deal with assets, it is an AI filter enhancing the rendered image. If it dealt with assets then it didn't need to work at run time, assets could be enhanced in advance.

Game publishers being giddy about it should ring alarm bells if anything.
Click to expand...

That may be. To be honest. Nvidia has said multiple conflicting things about this tech. Its lighting, its not a filter, it is a filter, It enhances geometry, it leaves models alone. I don't think they were actually prepared for any backlash. They seem pretty out of touch. Maybe when Jensen said 5 enhances geometry he is just wrong and no one there wants to correct him? I don't know lol

I may be wrong. As I understand what seems to be going on here. It sounds to me like ALL of this can be rendered by an external source. Hence the current tech demo has 2 GPUs... one to render the game, one to do the AI slopification. I don't see how if your streaming assets anyway you can't slop them at the datacenter as well. I mean if sonys patent works and you can play a 100mb game locally with GBs of assets streaming, why not?

We are for sure agreed on the money. When all the big game publishers start clicking their heals in unison its not good for us.
 
Niner said:
View attachment 792174
Click to expand...

Inaccurate, the color on the body should be orange, the eye lids should only be purple. Needs another 4 RTX 5090s to achieve proper coloration.

M76 said:
There is no prompting involved here, this is just a V2V generative AI like any other. The only things dev can control is how far they dial up the slop. What this might allow the devs to do later is to basically make a game of placeholder quality assets that the AI will try to pass off as faux realistic. It's the videogame version of death of the author as the game developers artists and designers have no precise control over the end result. Which will make every game look the same that uses the tech.
Click to expand...

I agree. I am not 100% against AI in assisting development. I am not against it helping make place holders, or helping clean up in some type of asset prior to review and adjustment. But DLSS 5 seems to be doing exactly what we don't want to see. It is essentially creating the end product without final adjustments from the artists. More or less allowing more sloppy work with what is essentially a crappy filter thrown on top of it. And like all upscaling and whatnot, there will be some downside. And we really don't need more of that on individual assets.

MikeTrike said:
DLSS is still lazy slop and creatively bankrupt. Nvidia is just trying to maintain a stranglehold over everything they can, and lazy developers are happy to ride shotgun to the bank. Although, hopefully it's off a cliff instead...

View attachment 792202


View attachment 792204
Click to expand...

Japanese and now seemingly Korean games put a lot more effort into characters. Resident Evil, Final Fantasy, Death Stranding, Stellar Blade, all games that generally have life like and realistic looking faces and interactions. Western game developers seem to be going more of the modern Assassin's Creed route lately. Even older Assassin's Creed games often had more realistic looking animations, better voice acting, and better facial animations than some of the newer AC games.
 
You must log in or register to reply here.
Back
Top