At its GTC 2026 event, Nvidia has revealed the next generation of DLSS. DLSS 5 isn't a frame-rate, frame generation or performance enhancing technology. Instead, it's Nvidia's attempt to use machine learning to leapfrog generations of GPU hardware evolution to deliver photo-realistic lighting using the hardware of today - and we saw it demonstrated in titles including Resident Evil Requiem, Hogwart's Legacy, Assassin's Creed Shadows, Oblivion Remastered and Starfield. The transformational lighting delivered by DLSS 5 is frankly astonishing and will be coming to RTX 50-series GPUs by "Fall 2026".
Plumbed into game engines in a similar way to DLSS super resolution and frame generation, the DLSS 5 lighting model uses just colour information and motion vectors to deliver photo-realistic imagery. Nvidia says the core goal is to enable developers to achieve the artistic vision desired by game creators limited by the technological limits of today's hardware. While lighting is radically revamped, all geometry, texture assets and materials remain as they were in the original game - but the effect can be astonishing.
The AI network powering DLSS is aware of the semantics of the scenes it processes. It "recognises" and processes elements like skin, hair, water and metal differently to apply photo-realistic lighting effects. While there could be some comparisons to generative AI, DLSS 5 is consistent and coherent in its rendering of the game world, the environments and the characters within it. It's capable of working with standard rasterised games, RT supported titles and path-traced experiences - the higher fidelity you give to the model, the better the end results you'll see from it in terms of material response, lighting and shading.
Character rendering is transformed. You'll see realistic subsurface scattering in skin and more realistic hair rendering. Games like Resident Evil Requiem, Hogwarts Legacy and especially Starfield show generational leaps in fidelity. Meanwhile, environments such as those seen in Assassin's Creed Shadows and Oblivion Remastered see a dramatic boost in the realism of subtle, realistic shadowing and ambient occlusion, grounding objects in a scene in a way the original renderer can't.
Handling of materials can be astonishing, with everything from metals, cloth and even the skin of fruits looking remarkably realistic. Particularly impressive is how DLSS 5 handles light and shadow around foliage - something that's very difficult for standard renderers to achieve, even with RT or path tracing. Right now, DLSS 5 is still a work-in-progress project - we did spot some screen-space errors, but Nvidia describes what we're seeing today as a "snapshot" of the technology as it stands, with further improvements and optimisations to come. It's set to be launched later in 2026 after three years of development at Nvidia.
From our perspective, it's important to see that Nvidia is showcasing DLSS 5 on actual shipping games (and its forward-looking Zorah demo) if only to demonstrate that the technology will only find use in cherry-picked scenarios. This is no Matrix Awakens demo - a tantalising glimpse of a future that still hasn't come to pass. DLSS 5 is the real deal.
There's much we don't know though. That starts with the computational cost of the ML algorithm. Nvidia actually used two RTX 5090s for its demos: one plays the game, the other exclusively runs the DLSS 5 technology. The use of two GPUs is required right now as DLSS 5 still has a long way to go in terms of optimisation - both in terms of performance and its VRAM footprint. However, DLSS 5 is designed for use on a single GPU and that's how it will ship later this year. Quite how scalable it is also remains to be seen, but in common with other DLSS technologies, Nvidia tells us that the computational cost scales with resolution.
Expect to see DLSS 5 as a further option with the graphics menus of PC games, alongside super resolution and frame generation, but the demos we saw were running with 2x frame-gen. In fact, DLSS 5 is integrated into frame-gen, which makes sense - after all, using this lighting technique, every frame is now generated. And yet, the quality is there, with few if any of the inconsistencies and mistakes typically seen within photo-realistic generated AI.
I'm fully anticipating a robust discussion around DLSS 5 and Nvidia's interpretation of photo-realism. In effect, the firm is using its advanced knowledge of machine learning to "bypass" the years of hardware evolution and software development required to deliver photo-realism by using ML to deliver its own vision for the future of graphics - but is Nvidia's interpretation of that future what gamers and developers actually want?
While DLSS 5 has its own model based on relatively limited game engine inputs, Nvidia says that there will be the means by which developers can interact with the technology to get the results they want. On top of that, DLSS 5 isn't a full replacement for current lighting - the algorithm requires game inputs to work and the higher the quality of those inputs (for example, path tracing over ray tracing) improves the end result of the DLSS 5 output.
And of course, if developers or indeed gamers don't like it, there's no compulsion for them to use it. Needless to say though, Nvidia says that feedback from developers has been positive and it already has a significant library of games with pledged support.
There's a lot to process here - and there's the sense that we're still not fully aware of the full implications of this technology - but the bottom line is that this is big. Bigger than the last big jump we saw in gaming graphics with the arrival of path tracing in triple-A games, kicking off with Cyberpunk 2077. And it highlights in the most dramatic way possible that innovation in graphics will come more and more from software rather than diminishing, generational leaps in pure hardware performance.
We'll have more on DLSS 5 soon, including a longer DF Direct on how we came to see the new technology, our immediate reaction to that Resident Evil Requiem comparison, and more detailed thoughts on each of the demos we saw. We're also looking forward to the reaction from our colleagues (in particular Alex!) and we'll be fielding questions about what we saw in the next Digital Foundry Q+A show. But in the meantime, I'd expect to see a range of DLSS 5 coverage from other outlets as Nvidia's GTC event progresses, and I'm very curious about the nature of the reaction the technology will receive from press, developers and gamers alike.





Comments 34
You guys have really been killing it lately with your new content. Great work!
Wow, that looks really, really bad. Girl looks like an entirely different person, like a botched plastic surgery. The others go into uncanny valley territory as well.
I'm sorry guys, but this looks absolutely awful. I cannot believe you are promoting this and I am thoroughly disgusted by this. I know Oliver has always been "Mr. GenAI" but this is just so disappointing.
I'll also add: Rich and Oliver saying that these look "closer to the original artistic vision" just rubs me the wrong way. This is shameful coverage.
What in the AI generated heck.
In some cases this is literally unbelievable, the Forrest in the assassins creed footage looks like real life on my tiny phone screen.
I’m not convinced about the faces yet, I think this has the potential to be generationally transformative, but I suspect it needs to be implemented carefully to not ends up looking AI sloppish.
The faces look completely different from how the ones the artists made. No thanks bro!
OMG the slop chases me into my games now? How do we escape this hellscape?
Hopefully, we will still be able to switch back to any version of DLSS in any game.
I refuse to believe you really think this looks good.
I can't believe you guys are actually praising this, it looks like an AI Slop filter for games.
I was always worried Oliver's obsession with GenAI would eventually lead to some sort of controversy but not like this.
the only one that arguably looks better are the starfield ones, because artless slop on top of artless slop is kind of a wash, but even those they are pretty radical changes to the character design, as uninspiring as todd howard's latest wild ride may be. the rest are absolutely loathsome, and all characters look like different people with DLSS5 on, with different vibes, and i hate it.
can't wait for kirkification of all video games under herr huang's reign
Odd that the "enhanced" version of Grace standing in the street greatly resembles Vanessa Kirby, but in the other "enhanced" screenshot she looks completely different.
I'm not sure what is more worrying, the lack of consistency, the AI slop aesthetic, or DLSS potentially stealing the likeness of actual people and the inevitable legal sh*tshow that spills over to developers and publishers.
No. Grace literally has different lips between the two comparison shots. Either I am too used to baked in old school lighting or some generative AI is being used here to Yassify the models to make them look like what you see AI generates these days when input for "generic good looking woman" comes up with. I do not like how this first look at DLSS5 looks like. Not one bit.
Honestly, I've followed and respected you guys forever.
Supporting this will kill your reputation.
Incredible that you guys clearly don't see the problem here, omg the disappointment hearing you talk about how "impressive" turning the faces that had a previous art direction to a slop prompt: "hey grok make this character ultrarealistic" can't believe it. Please read the comments and acknowledge you are really out of touch.
Wow, this "AI Slop"-mode looks absolutely horrible.
Hey guys, made an account JUST to come say this sucks. We need to be full throated about our opinions on generative AI, or it’ll continue to seep into everything. Public opinion is the only thing that can realistically slow this down - or the bubble pops, whichever comes first.
What did they do to hot unc' Leon? I can see some aspects of the image being better but it kills artistic intent and it does constantly remind you that you are looking at generative AI which is an uncomfortable feeling and really weird. Like the dlss off is a grounded image and dlss on you're hallucinating on drugs.
Yeah I can't get behind this at all. Almost everything here looks significantly worse with DLSS 5 turned on - its so obviously destroying the original artists' intent, lighting errors all over the place, just a complete nonstarter from me. They just look like Snapchat filters come to life.
Thanks for letting us know about this but wow this is so bad. I get you guys are impressed by the tech but surprised you're saying this looks good.
I don't see how any artist developer would be in support of this either.
With you mentioning this being injected into games that don't support it, could we see less games on PC if this is possible? Only shipping on console to avoid this happening to their artistic vision.
Made an account just to comment here. This is really shameful coverage of DLSS 5. The video and this article fail to mention the many many many implications of this technology being used. You say "oh it only changes the lighting" It really doesn't. Its inherently changing the artistic and creative integrity of games seemingly just as a business decision. It's an AI slop filter and that's all it is. This video is also just a blatant ad too.
Love you guys, but I don't know how you can say with a straight face that this is just lighting. It obviously is not. It's AI adding texture, shading, and color to create something completely different from the original artwork. It's awful, and it is the pure unfiltered definition of "AI slop." For crying out loud, it turned a teenager into 40-year-old man in the shot from Hogwarts Legacy near the end of the video!
This is awesome and clearly the future of gaming. People here are complaining about its looks when people did the exact same thing with DLSS 1. Now its many times better. this is the first appearance of DLSS5's neural rendering. It's going to get much better.
As you can see, my young apprentice, your rasterization has failed. Now witness the firepower of this fully ARMED and OPERATIONAL DLSS station!
@Samuel-Gamer when you say supporting DLSS 5 will kill their reputation, what do you mean by "supporting" it exactly?
Writing an article on it? Recognizing anything positive at all?
Despite all the hate out there I would choose the dlss 5 version of starfield over the original version every day of the week. And i bet most people will do the same in the end. Also if you prefer the old poly characters just dont turn on dlss 5? Why are people so upset. Options are nice to have.
The way it transforms Grace's face into generic "beauty" (complete with changing the shape of her eyes and mouth, turning her blond hair into highlights with darker roots, and adding heavy make-up), while leaving Leon still mostly looking like himself, is consistent with the well documented gender bias of gen AI, and its datasets.
DF, I'm not so sure this is rendering. Rather, this is a deepfake. If it's using generative AI, which almost certainly seems to be the case given its virtually identical resemblance, then not only is it painting over the original artistic intent, it's stealing from other artists to do so.
I would've expected a more critical appraisal from the DF crew for something this transparently controversial. Hopefully your follow-up discussions can shed more insight on this. This tech should be scrutinized, not accepted uncritically.
I needed to make an account just to say that it looks awful and I'm actively against this horrible AI grok yassification filter
I genuinely thought this was a joke when I first saw the pics posted online. Absolute unbelievable dogshit and an insult to artistic intent. What a waste of talent and money by nvidia
I love how Prof Hecat from Hogwarts Legacy, who was once aged by being "wounded by time itself", is yet again aged thanks to nvidia this time. almost all faces look like they receive movie poster level dramatic lighting regardless of the environment. ofc nvidia would want us to shell out for two 5090s but it doesn't mean that it's good for the gamers or the studios, or even exciting if I'm honest.
Just sad to see this, and also afraid of the future of the already depressing AAA industry.
This is obviously beyond lighting. This is an aesthetics overhaul.
If you think this looks like a logical "step-forward" for gaming visuals, then you have to be a fan of generative AI imagery and its general appearance, or maybe too numb to recognize it.
These characters don't need AI face-filters adding makeup, fillers, dye-jobs, and a general uncanny look that doesn't match the artists original input or intent. This looks closer to a mod. NVIDIA tech changing the artists and developers aesthetics and design to "fix" their games is a bizarre thing to be excited about.
I am honestly confused how DF can't spot this uncanny, AI face-filter look a mile away and trying to pass this off as a normal thing. Its goofy looking stuff that, unfortunately, a lot of people will enjoy because a lot of people enjoy generative AI imagery slop.
Don't want to jump on the bandwagon but these genuinely look out of place. I wish they would at least have some more cartoony/non realism examples because all these examples look like the AI just guessed what a face should look like based on training data instead of increasing the clarity of what's actually there.
Guys, stick to the technical stuff. The fact you didn't see the audacity here and slap in the face to art directors / character designers /animators /concept artists etc who made these games, is shocking. But I get it. You're all nerds who don't create art or really understand what goes into it. Really embarrassing.
Endless praise and zero pushback.
Not even a tiny bit, not even questioning the premise of "anticipating the artistic vision of game developers" and what that a truly means in the context of over half of the GDC 2026 State of the Game Industry survey participants stating that they think generative AI is harmful to this industry.
You parrot Nvidia's PR line about "positive feedback" from developers but did you actually ask any of them at GDC for their opinions on this? Because they already seem to have a firm stance.
You did not even try. This might as well be an article on Nvidia's site. This is not independent journalism.
Also, here's a fun question: Did Capcom, Ubisoft and the other publishers that own the rights to the games in this DLSS5 presentation look at these results beforehand and think it was a good idea for their brands to have Nvidia publish these modified assets to be mocked online? To have the creative endeavors of their employees (which are the results of many years of work) trampled on by this AI filter? Do you think any of the people that worked on the Grace or Leon models (or any other character models featured here, or the very detailed background assets) looked at this and said "Yes, this is obviously what I had in mind"? Do the staff of Digital Foundry truly believe that? Seriously?
Show Comments
Leave A Comment
Hold on there, you need to login to post a comment...