Nvidia brings AI slopification to games with DLSS 5.

  • Thread starter Thread starter namename11
  • Start date Start date
  • Views Views 2,233
  • Replies Replies 39
  • Likes Likes 6
I don't get the hate. It's entirely optional, and it DOES improve the graphics in many occasions. Specially environment.
People are focusing too much of the faces of the 3D character models, which admittedly can be worse and should be made optional somehow, since it's not always an improvement.

Here's hoping the tech gets even better.
 
  • Like
Reactions: Skelletonike
They used two 5090s to showcase this. I don't imagine anyone in their hands got two 5090s in their PCs. The idea of this alone makes me wonder the point of high-end gaming and stuff like that
This was very much a "look what we can do even though you can't" kind of thing.
 
This was very much a "look what we can do even though you can't" kind of thing.
Makes sense. I wouldn't want to even use this since I believe that having something like this would make people not have to worry about putting too much effort in the visuals for their games since they can use the AI to make more use of them.
 
I don't get the hate. It's entirely optional, and it DOES improve the graphics in many occasions. Specially environment.
People are focusing too much of the faces of the 3D character models, which admittedly can be worse and should be made optional somehow, since it's not always an improvement.

Here's hoping the tech gets even better.
Optional, sure, but the fact it's being advertised as a major part of the next DLSS, if not "the" main feature of it, doesn't hit people very well for what is essentially a filter. And what does it have to do with DLSS anyways? Honestly, nothing. Deep Learning Super Sampling. Nvidia already muddied it up with this like frame generation and ray reconstruction, but the way I see it, those are meant to to retain the vision the developers had while not requiring excessive hardware to accomplish them. Upscaling via DLSS allows a low-end GPU, like the one in the Switch 2, to produce an approximation of something the GPU using only shaders could not do in a playable form. But this example of DLSS5 was done using 2x 5090 GPUs, something pretty much nobody has, nor can afford. Why not just 1x 5090? Does it really require that much hardware? If so, then who is it for?

My problem with this is that it seems to only be a means to continue pushing AI, and not whether it's an actual good idea or progresses the point. Reiterating what I already said, you look at the various iterations of DLSS, and you see that they all (or at least most of them) are there to grant devs the means to do something that would otherwise be unplayable with current hardware in a raw fashion, retaining their vision. You can only go so low in resolution and/or frame rate before folks get turned off by it, so this tool allows the ability to make them more playable. But this "AI slop" of an iteration, imo, seems to have no real ties to that. So why are they even doing this?

Nvidia is in the lead with AI, and are so deep into it that they have to find something..... anything to keep up the momentum.
Post automatically merged:

The last time Nvidia simply took an image and altered it was back when they first created DLSS, processing a frame with spatial upscaling. And boy was that terrible. It wasn't until DLSS 2 where they shifted to temporal upscaling, which required more inputs like motion vectors in order to produce an output that was actually pleasant to see. With DLSS 5, they basically jumped right back into screenshot processing mode with nothing else (looked into it, and they are using motion vectors, but that may simply be for the upscale portion) and gunked it up. No other inputs produced by the scene itself is being utilized. No lighting information, no geometry information, none of that. So it really has no clue about the scene in depth.
 
Last edited by DiscostewSM,
  • Like
Reactions: AlexMCS
I don't get the hate. It's entirely optional, and it DOES improve the graphics in many occasions. Specially environment.
People are focusing too much of the faces of the 3D character models, which admittedly can be worse and should be made optional somehow, since it's not always an improvement.

Here's hoping the tech gets even better.
It's a filter. Nvidia employees even admitted you cannot just turn it off for certain objects. You can only apply it to the whole already rendered frame. You get a few sliders that adjust the intensity, but that's pretty much it.
 
  • Like
Reactions: Xzi and AlexMCS
The name is indeed terrible (nothing about SS here), and since it's a frame wide filter, yeah, not liking the way it works.
I like AI generated frames and scaling to a certain extent, but simple AI-based filters are a pass.
Too bad, since I liked what it does to some of the environment scenes.
 
The big problem with all of this is that the model is trained on so much data, that it is making approximations on details that really aren't there. Like, say you have NPC A. The filter isn't designed to make approximation based only on NPC A, but the thousands and more NPCs that could have similarities to NPC A. So all of a sudden, they are getting details that don't exist in the original character model, thus becoming the "AI slop" we are seeing in the preview, from additional patches of hair, skin that appears more shiny in a lighted scene, etc.

I can see a path forward with this though, but not with Nvidia handling the training, because they want it trained in anything and everything. Instead, the devs of each game would be in charge of the AI model to train it on higher-quality versions of the characters/objects/etc they themselves are making. The model, when enabled in the game itself, would have better approximations because it isn't having to go through all the "slop" of all that is irrelevant. It would be on a per-game basis unfortunately, but I would also imagine the model would be smaller and faster in the process. At least, that is how I'm looking a this.
 
According to the CEO himself it is absolutely not a post processing filter, and does indeed apply at the geometry level, with full artistic control to the developer. Whether you believe him or not is up to you. He did claim 5070 had 4090 performance so there's that.
 
According to the CEO himself it is absolutely not a post processing filter, and does indeed apply at the geometry level, with full artistic control to the developer. Whether you believe him or not is up to you. He did claim 5070 had 4090 performance so there's that.
And yet the folks developing the feature are saying....

Screenshot 2026-03-25 003317.png



 
It's entirely optional
I'd agree if Nvidia was calling this one individual feature of DLSS 5, but they're saying this is DLSS 5 in its entirety. Given their GPU market share and how much the US economy is dependent on them, they can push the gaming industry in any direction they want. That said, if nobody can afford/cares enough to buy the hardware needed to run it on the consumer side, developers are likely gonna be hesitant or very slow to adopt it from their end it as well.

it DOES improve the graphics in many occasions. Specially environment.
I very much disagree, it ignores the context/weather of the environment, cranks up contrast and puts makeup studio halo lighting on everything. If this is the future of AAA gaming, indie games are gonna become even more dominant than they already are.
 
Last edited by Xzi,
If this is the future of AAA gaming, indie games are gonna become even more dominant than they already are.
I have already corrected my stance since then, but this is a very good thing IMO...
"AAA" gaming has been a disappointment to me for years, with very few exceptions.

That said, I still think it improves the environment and I disagree with your assessment of "halo makeup", but if that's the extent of it (no optional toggles, no actual super sampling and scaling, and the huge computational cost), it sucks.
 
Let me get this straight: We now have three options when it comes to the latest AAA titles:
1 - We can play them stuttering with low resolution (DLSS off)
2 - We can play them at acceptable resolution with an acceptable framerate and ghosting (DLSS 4)
3 - We can uglify them for the cost of a SECOND RTX 5090.

I still hope we'll see better optimized games due to high hardware prices... at least for a year or so.
i'm insanely nitpicky about ghosting and visual artifacting, dlss 4.5 is terrifying, it doesn't have any if not very very minimal ghosting, significantly less than default TAA with native res
 
I have already corrected my stance since then, but this is a very good thing IMO...
"AAA" gaming has been a disappointment to me for years, with very few exceptions.
For AAA games that are a disappointment due to bad gameplay, yassified faces aren't gonna fix that. For AAA games with strong gameplay and a specific vision for art direction, this filters out their identity in favor of something much more generic.

I would say the one practical application of this is breathing new life into older AAA games, but with how stiff the animations are, the uncanny valley would be impossible to ignore. And that's really the biggest problem: it looks terrible in motion, the AI completely fails to keep geometry consistent across different camera angles. Even in Nvidia's own demo, Grace has two different faces depending on whether she's looking straight at the camera or it's at her side.
 
  • Like
Reactions: AlexMCS
For AAA games that are a disappointment due to bad gameplay, yassified faces aren't gonna fix that. For AAA games with strong gameplay and a specific vision for art direction, this filters out their identity in favor of something much more generic.

I would say the one practical application of this is breathing new life into older AAA games, but with how stiff the animations are, the uncanny valley would be impossible to ignore. And that's really the biggest problem: it looks terrible in motion, the AI completely fails to keep geometry consistent across different camera angles. Even in Nvidia's own demo, Grace has two different faces depending on whether she's looking straight at the camera or it's at her side.
Just to be clear the quoted "this" refers to "indie games becoming more dominant".
I don't really care for AAA games anymore.

DLSS5 isn't as good as I thought it could be, after further research, and the required computational power is hilariously prohibitive.
 
The other possibility is that devs just use this Ai for an idea tool and change their actual models. No need for people at home to actually run the Ai.
 

Site & Scene News

Popular threads in this forum