Nvidia DLSS 5 first look: generative AI lighting radically transforms game visuals

Daniel Sims

Posts: 2,442   +74
Staff
First look: Nvidia CEO Jensen Huang used his GTC 2026 keynote to unveil the next stage of machine learning-driven video game graphics, which substantially transforms lighting and the overall visuals of a game using generative AI. Although the company's demonstration showcased dramatic increases in detail in titles such as Resident Evil Requiem and Starfield, it immediately drew sharp reactions from users, many of whom compared the results to AI-generated video footage.

Instead of upscaling games from lower resolutions or interpolating frames using AI, DLSS 5 applies machine learning to a game's lighting model. Nvidia calls it the next stage of rendering after upscaling and ray tracing. Digital Foundry got an early hands-on look at the technology (video below), which sparked controversy within hours of its debut.

The changes DLSS 5 introduces to environments and characters are so drastic that some social media reactions accuse Nvidia of effectively layering an AI filter over developers' original artwork. That perception is reinforced by the fact that several of the demo images resemble the glossy, hyper-processed aesthetic often associated with AI-generated photos and video.

Nvidia stresses that DLSS 5 does not replace models or textures. All changes result from newly generated lighting based on a game's color and motion vectors, substantially altering subsurface scattering, material interactions, and other rendering processes.

The company's press release also states that the technology can handle materials, though it does not specify whether the AI generates entirely new material properties or simply adjusts existing ones. Huang described the system as applying generative AI to lighting, but the process remains deterministic and tied to the underlying game data, meaning it does not hallucinate new objects or geometry.

The update is already sparking discussions about authorial intent and compatibility with various art styles. Earlier versions of DLSS reconstructed images based on high-resolution training data to preserve each game's artistic intent. In contrast, DLSS 5 may risk taking some of that artistic intent out of developers' hands.

Nvidia frames the goal as pushing toward photorealism, comparing DLSS 5's output to Hollywood-grade visual effects. In the company's demos, the approach appears to work well on realistic characters in games like Resident Evil Requiem and EA Sports FC. But applying the same system to more stylized or fantastical worlds – such as those in The Elder Scrolls IV: Oblivion, Starfield, or Hogwarts Legacy – could plunge visuals into the uncanny valley.

Developers can tune DLSS 5's intensity, color grade, and masking. They can also exclude certain elements from the technology's effects, but the extent of artistic controls remains unclear.

Despite the early criticism, numerous developers have approved DLSS 5 and collaborated with Nvidia to create the initial demonstrations, including Bethesda, Capcom, and Ubisoft. Assassin's Creed Shadows, Resident Evil Requiem, EA Sports FC, Hogwarts Legacy, The Elder Scrolls IV: Oblivion Remastered, and many other titles will be supported.

DLSS 5 is expected to arrive this fall with support for RTX 50 series graphics cards. Yet one of the biggest unanswered questions is computational cost. These early Nvidia demos ran on dual RTX 5090 systems, with one GPU dedicated entirely to DLSS 5 processing. The shipping version is expected to run on a single graphics card, but we'll have to wait and see if that's truly feasible.

***

What follows are some of the most interesting and funny X reactions we've seen so far. Make sure you leave your comments below:

Permalink to story:

 
I think for enviroments it could do wonders. For characters - some examples were good, others..not so great.

I have a feeling that the 5090 will be the only card capable of running it at a decent framerate - Them running Dual 5090's means they're far off the target at the moment
 
It is impressive but on a grand scale it is... Almost worthless. What am I expected to do, admire this feature when the game has cutscenes? Does anyone even care? Heck, even body jiggle physics are more important because that at least is visible at all times in 3rd person games.
They should focus on optimization, both game devs and Nvidia with its drivers.
In this economy, the majority of gamers are doomed to stay forever on their ancient 1060 GTX or rtx 3060.
 
AI video inspired by the game graphics is a weird way to go.

Also, wouldn’t that take a giant amount of data to generate and compute to stay real time? I can see why they had a 5090 just to run the DLSS effect.

Maybe when they cram this into a 5060 8GB the reduced effect is subtle enough to not be weird but that’s a rather strange thing to hope for.
 
I noticed for me dlss 4.5 model m had some hallucinations in details that weren't their before and had a feeling Nvidia would be going this direction with never models.
Most gamers just want a better more efficient anti aliasing technique and improved path tracing performance. I hope they have a slider and or toggle which dlss 5 features you want like facial and character overhaul/ environmental overhaul as well as how impactful you want that filter or not.
The fact that they needed 2 x 5090s what sums up to around $ 6000 worth of graphics cards and a 1200 watts of power potentially just to run this is crazy. Allegedly they can run this on one card, but who knows how that scales. Does anyone feel like Nvidia raised the gama levels to make the image stand out?
Imo it has potential on older titles in our libraries. Who needs to pay for endless remasters when you can add dlss 5 filter unless that's what developers will do as well and charge us for it going forward 🤔.
We thought hit the surface with ai slop; we haven't even reached the 1st layer. 🤪
 
Well, it does look different, but not necessarily better. I guess it may need more tweaking.

Some demos look great, others so so and a couple look better without it.

BTW IIRC the SW RTX demo initially ran on 4 Pascal Titan cards at 1080p, then the RTX2080 came out and single handedly ran it at 4k, so there's that.

 
Well, it does look different, but not necessarily better. I guess it may need more tweaking.

Some demos look great, others so so and a couple look better without it.

BTW IIRC the SW RTX demo initially ran on 4 Pascal Titan cards at 1080p, then the RTX2080 came out and single handedly ran it at 4k, so there's that.
It's true but Nvidia killed off sli and now they reached a ceiling so they are reneging back.
 
It looks terrible, especially on people, like some real time Instagram filter.

Unsurprisingly DF are praising it like they would anything from Nvidia, they're pretty much unofficial Nvidia marketing.
Essentially if a reviewer refuses to cooperate with Nvidia, they will be on their blacklist. So unsurprisingly, most reviewers are praising it. Nvidia is no longer bothered about gaming as we all know. This DLSS 5 is not progress in gaming but more a side step into more AI. DLSS was first introduce to help increase frame rates without sacrificing too much visually. Now you need a RTX 5090 to have AI paste someone's face on the in game characters is not about gaming at all.
 
Remember that rumored dedicated AI card? This is what it's for. 100% their strategy is to sell a second GPU or a dedicated card for this. The demo needs dual 5090 and in a few months it's going to run on a single 4070? Heck no. They do not release cool things for free (not that I think this is cool).
 
This seems like a very interesting artistic addition to the current gameplay.
B9v0pyy.jpeg
 
I'll just wait for oh the 3rd or so iteration of AMD's copy of this technology, which should come about in just over 4 years give or take.

By which point everyone will agree that it's awesome.
 
I'll just wait for oh the 3rd or so iteration of AMD's copy of this technology, which should come about in just over 4 years give or take.

By which point everyone will agree that it's awesome.
By that point game devs might be able to make this not look awful.

However with UE5 games being un-optimized garbage, and now AI completely changing game visuals for the worse, Nvidia is doing a great job at saving me money by not buying new games,lol.
 
Developers say:
"We are against it because it changes the characters' intentional creative design by the developers"

So does that mean the developers "intentional creative design" was to create something with no facial feature details and dead like animation?

I see nothing wrong on those examples provided by NVIDIA. It adds detail and depth to the characters faces.
 
Last edited:
Developers say:
"We against it because it changes the characters' intentional creative by the developers"

So does that mean the developers "intentional creative design" was to create something with no facial feature details and dead like animation?

I see nothing wrong on those examples provided by NVIDIA. It adds detail and depth to the characters faces.
I can understand developers being against this because it's akin to taking someone's art and running it through some gen AI prompt, completely changing the original art.
It should be up to the original game devs to decide if they want hyper-realistic faces in games, not being forced by a GPU company.
Faces look way too artificial and plastic with DLSS5 on, and another concern is this won't be optional at some point because every developer wants to implement those features.
 
Back