It’s been a frustrating few days as someone who saw Nvidia’s DLSS 5 demonstration at Gamescom last week. Since a recent mod leaked that showed DLSS 5 adjusting some characters in a pretty confusing manner, I’ve been under embargo and unable to discuss the true functionality, or intent, of the AI-powered filter.
First teased back in March 2026, the feature received a largely negative reception due to the stark differences in character models, with some claiming they were “unrecognizable.” To address the elephant in the room–yes, some of the DLSS 5-infused creations are horrific. Final Fantasy’s Cloud looks like something straight out of a horror game, while adding multiple layers of DLSS 5 has made characters from The Legend of Heroes series look utterly harrowing. But from my time with Nvidia at Gamescom, I’m trying to be realistic about the use and power of the feature.
“DLSS 5 is intended to be used for photorealistic situations and characters,” Edward Liu, Nvidia’s director of applied deep learning research said at the demo. That means no Final Fantasy, no Kingdom Hearts, and nothing that would stylistically suffer as a result of implementing it. What’s more, developers can opt in or out of using the AI-powered tool, and there’s more customization than just “on” or “off.”
For example, developers can create masks in their games to ensure DLSS 5 is only applied to those areas. If you just want to enhance the appearance of food in a game, that’s possible. If you’re just looking to make the foliage and plants look better, that’s okay too. The intensity of the filter is completely customizable by the developer. Perhaps more importantly, it can be disabled entirely as a gamer–there’ll be a simple toggle in the menu to turn it on or off.
DLSS 5 uses several layers to preserve artistic intent. It’s trained to recognize the surface albedo, lighting semantics, and surface normal. In simpler terms, it looks for the original measure of how much sunlight hits the surface, the way the lighting is designed, and the direction the surfaces are facing to determine the way DLSS 5 is applied and ensures it’s done in a way that does not change those three original layers.
It’s important to remember that DLSS 5 is something the developer opts into. As such, it’s highly unlikely any of them will be deliberately implementing a tool that creates an eldritch horror instead of making their characters, objects, or environments look better.
The first game to officially launch with DLSS 5 capability will be NBA 2K27, where photorealism is obviously the intent. As such, character models appear more detailed, and even vast numbers of NPCs in the crowd can easily have DLSS 5 applied to improve their appearance.
There have also been reported concerns around performance. Nvidia states that, when using its 50-series GPUs, DLSS 5 has a minimal impact on performance. When running NBA 2K27 on the Ultra settings at 1080p using 6x Multi Frame Gen, a 9800X3D, and 64GB of RAM, the 5090 runs at an average of 797 frames per second. The 5080 follows up at 602, with the 5070 Ti at 530 and 5070 at 386. The 5060 Ti is at 325 and the 5060 is 258.
Of course, with the average person’s setup not quite rivaling that of Nvidia’s, there will be variations to that. However, we were shown various builds running NBA 2K27 on different settings, from high-end to mid-range, with next to no performance issues, though we haven’t yet been able to independently verify those figures.
For me personally, whether I choose to toggle DLSS 5 on or off will depend on the developer’s use of the function. If egregious filtering is used to the point where I no longer recognize the characters, or I find myself distracted by the filtering, I’ll turn it off. But if Nvidia’s demonstration is accurate to the level of customization it provides developers and preserves artistic intent, I don’t see any reason I won’t be using it.
