NVIDIA’s “Significant Breakthrough” In DLSS 5 Is Looking Like An AI Slop Filter For Games

Just a day before GTC, NVIDIA was being particularly bullish about the graphics technology they’re about to announce in the event, claiming it as “the future of real-time rendering“. The tech in question? A new version of DLSS, aptly named DLSS 5. Except it’s not the future that gamers at large wanted, going by community’s reactions.

DLSS 5 “Slopifies” Graphics

In NVIDIA’s words, DLSS 5 features a new “real-time neural rendering model that infuses pixels with photoreal lighting and materials,” claiming photorealistic graphics “previously only achieved in Hollywood visual effects.” NVIDIA CEO Jensen Huang even went as far to claim that this new tech is “the GPT moment for graphics.” The result demonstrated in the video, however, is what you may describe as uncanny valley, or straight up “slopified”. (We’ve even seen the word “yassified” brought up in reference towards RE9 protagonist Grace’s “new” look.)

The idea behind DLSS 5 is to essentially skip through some of the rendering steps in compute-intensive path tracing and let AI figure out how the lighting and textures should look based on the scene itself. In other words, path tracing can’t compute the lighting fast enough (the company is keen to point out Hollywood films require hours to render each path-traced frame for the sake of photorealism, as opposed to just 16 milliseconds per frame available to games), so AI comes in and takes care of it, using data like a game’s color and motion vectors combined with its trained model.

Here are some of the examples provided, and for starters, Grace indeed looks different, but perhaps not in a good way. There’s worse examples – the Hogwarts Legacy example shows the old woman with a completely different look with telltale signs of generative AI infused into the result, like excessive contrast in lighting and overly pronounced details (basically like those stock images you see if you have ever search it up on Google). Starfield is another one with extreme changes, though in this case the game’s suboptimal facial graphics is partly to blame.

Seemingly in response of the backlash against DLSS 5, NVIDIA even left a comment under its YouTube video that goes:

Important to note with this technology advance – game developers have full, detailed artistic control over DLSS 5’s effects to ensure they maintain their game’s unique aesthetic. The SDK includes things like intensity, color grading and masking off places where the effect shouldn’t be applied. It’s not a filter – DLSS 5 inputs the game’s color and motion vectors for each frame into the model, anchoring the output in the source 3D content.

The glass-half-full takeaway from the statement above is that developers will make sure that whatever comes out of this tech is the intended artistic direction, but as some examples have been presented by this point, even technologies like Multi-Frame Generation and Super Resolution are being abused by some games as a “crutch” to make sure the game is at playable framerates, effectively making them mandatory features. Can’t blame the gamers out there from taking the glass-half-empty approach at this one, could you?

In its FAQ, NVIDIA says the feature will be released sometime in fall 2026, though exactly how the performance requirements is going to look like is yet to be clear. NVIDIA further noted that it uses two RTX 5090 cards to showcase this demo, with one running the game’s graphics while the other is exclusively responsible for DLSS 5 processing (as it requires 22GB of VRAM right now). While the company assures that optimization will make it work on a single GPU, it’s hard to say if it’ll work on cards with less VRAM capacity available, like the RTX 5060 8GB, given the AI model’s inherent thirst for memory.

Pokdepinion: Some of these examples definitely qualifies for the uncanny valley.