DLSS 5 (and the Backlash) Explained

[00:00:00]For 8 years, Nvidia's DLSS technology made your games run faster with almost no downside. The company caught some flak for inflating frame rate numbers with AI-generated frames, but generally DLSS had a good reputation. That is, until early 2026, when DLSS 5 shattered it in the eyes of anyone who'd prefer that their games not resemble the AI slop infecting the rest of the internet.

[00:00:26]But is that backlash justified? How different is it really from its predecessors? Nvidia's original deep learning super sampling launched with the RTX 2080 in 2018. And while this is a gross oversimplification, it essentially rendered frames at a lower resolution, then upscaled them using machine learning techniques to get a sharper image without the performance hit. Later versions refined this basic image enhancement process with better anti-aliasing and denoising. But DLSS 3 added frame generation, a new kind of interpolation that used AI to boost frame rate by generating what many gamers derisively call fake frames. The current version, DLSS 4.5, can be set to generate up to six fake frames for every real frame. That is, a frame that is, although altered, still created by traditional graphics rendering techniques. DLSS 5 seems to take this fake frames concept even further and gets rid of real frames altogether.

[00:01:27]According to Nvidia, DLSS 5 uses a real-time neural rendering model that infuses pixels with photoreal lighting and materials. So, skin gives off that soft subsurface glow, fabric gets a proper sheen instead of looking flat, hair stops looking like a plastic helmet glued to someone's head. And the demo of Nvidia's Zora environment genuinely looks stunning with forest overgrowth and bouncing light that would melt the average GPU if rendered the traditional way. I mean, rendering it at all would probably melt anything other than the two RTX 5090s they used to run these demos, which is another unusual thing about DLSS 5. At least for now, it seems like the new DLSS isn't about performance at all. It's purely about visuals. But, whether you think those visuals look like slop or not, the tech is real. The question is how it works.

[00:02:20]Nvidia CEO Jensen Huang has said DLSS 5 operates at the geometry level, and that developers will have full creative control over the model's output. And that makes it sound like it's plugged deep into the game engine, working with developer-created assets. But, is that actually true, Jensen? Really?

[00:02:38]Jensen? Cuz we're going to find out after we thank this video's sponsor, Micro Center. They have their laptop savings event going on this month.

[00:02:47]There's great discounts, including some devices sporting RTX 5070 TIs, like the ASUS ROG Strix G16. Austin, Texas is also getting their own Micro Center soon. To sign up for the grand opening and get a free 128 gig flash drive. And the Columbus location is currently being remodeled, but you can still pop in and grab a flash drive there, too. Finally, for all things Micro Center and tech-related, stay up to date on Micro Center news using our links below. It turns out that Jensen's grasp on the truth didn't quite operate at the geometry level. After the DLSS 5 reveal, YouTuber Daniel Owen got some real answers from Nvidia's Jacob Freeman about how the tech works. Freeman confirmed that DLSS 5 takes a 2D frame and motion vectors, i.e., information about the direction things are moving, as input. [music] Nothing else. It doesn't have real-time 3D knowledge of the game's current geometry, texture files, or other assets. It infers everything from the finished image, which is a fancy way of saying it takes a 2D screenshot of your game as input and uses generative AI to output a more Instagram-worthy image. No matter what you think about the AI look so clearly displayed in Nvidia's demos, some of which, especially shots of environments without people in them, looked practically photorealistic, there is some genuinely impressive tech here. How impressed gamers are by that tech seems to depend on how they feel about an AI model overriding creative decisions about what a game and its characters look like. Because the main problem gamers had with the DLSS 5 demos is how character faces in Resident Evil Requiem, Starfield, and Hogwarts Legacy were drastically changed. Resident Evil's Grace Ashcroft gained makeup that wasn't there, a Starfield character grew hair out of nowhere, faces looked airbrushed, and the word "yassified" started trending. Former Red Dead Redemption 2 developer Mike York called it a complete AI rerender, not a lighting pass, which, given what we know about how the tech works, is accurate.

[00:04:49]Now, to be fair, an Nvidia FAQ says developers aren't completely powerless to shape how DLSS 5 renders their game.

[00:04:57]It says they can adjust blending, contrast, saturation, and gamma, and mask elements out, so the AI skips over those. Essentially, they can do some basic Lightroom edits. But the AI still decides what to change everywhere else, and it's deciding from, once again, a simple 2D screenshot. Now, DLSS 5 doesn't ship until fall with launch titles like Assassin's Creed Shadows, Resident Evil Requiem, and Oblivion Remastered. So, there is time to give artists more control and stop the model from adding makeup to characters who weren't wearing any. After all, the first DLSS had a bit of a rough debut in 2018 before becoming the gold standard for upscaling. So, maybe this video will age poorly and people will learn to like DLSS 5, but we I that will depend on whether Nvidia spends that time addressing gamers' concerns or just insisting the people raising them are wrong. And all the neural rendering in the world doesn't matter if your screen can't keep up. The display is the final stop between Nvidia's AI and your eyeballs. And OLED is the panel tech that's quietly become the standard for making any of this look as good as it should. So, if you want to know what makes OLED so good for gaming and whether it's worth the upgrade, our video on OLED displays will tell you everything you need to know.