Sign in to use highlight and note-taking features for a better reading experience. Sign in now
Nvidia announced a new version of its DLSS AI upscaling technology for its graphics cards earlier this week at its GPU Technology Conference (GTC), which it calls the Super Bowl of AI. But unlike previous versions of DLSS that used AI to improve frame rates in video games, DLSS 5 has a much more ambitious calling: using generative AI to make character faces in games look more realistic and detailed. The demonstration received sharp blowback on social media, with many finding the effect off-putting, reacting with outright disgust, and calling it yet another example of AI slop.
DLSS, or deep-learning super-sampling, is a feature Nvidia introduced on its graphics cards in 2018. The primary use has been to improve frame rates in video games by rendering games at a lower resolution, then using AI to upscale the quality. More recent versions of DLSS insert AI-generated frames in between actual rendered frames. These techniques use less computing power than generating the full frames, allowing for better gaming performance without taxing your PC’s hardware and maintaining visual fidelity. The feature can be turned on or off.
“From a technical standpoint, it's quite an achievement,” Kevin Bates, CEO and creator of the open source retro gaming handheld Arduboy, wrote in a message to WIRED. “I would have expected a cloud-based rendering service to provide it. The fact they expect to distill it down to what can run on a single [graphics] card later this year is insane.”
But DLSS 5 has crossed a generative-AI rubicon. Instead of just being a tool Nvidia provides developers, it manifests as actual visual changes without their consent. While you can still turn it on or off in your video games, the technology has some developers—not just gamers—worried.
Nvidia showed off a demo of the tech on games like Capcom’s _Resident Evil Requiem_, Ubisoft’s _Assassin's Creed_, and Bethesda’s _Starfield_. The company says it's meant to improve the graphics and generate photorealistic details and lighting. The demo seemed to largely improve the lighting, which detractors compared to the glow of a ring light just out of frame. Faces became far more detailed, even introducing new facial features. It was also criticized on social media for over sexualizing characters, where people called the look “yassified,” or “porn faces,” and compared the effect to Instagram or Snapchat’s glamour filters, which smooth out imperfections on a person’s image. Gamers did not approve. _The Verge_ called it motion smoothing, but worse.
The tech also has other issues, like introducing unexpected artifacts in real time. You can see some of those problems in the official demo video itself. In a scene in a FIFA game where a soccer ball is being kicked into a net, the ball has weird artifacts on it with DLSS 5 on, looking like a piece of the net is on the foreground of the ball before it has even gone in the goal. (Pause the video at 59 seconds.) People’s facial features, like the female character in _Resident Evil Requiem_, have some slight but noticeably different facial features: larger eyes, fuller lips, and a completely different nose.
“It devalues an artist’s creativity and intent on a basic level,” says James Brady, a video game artist and designer who has worked on games like _Call of Duty: Modern Warfare 3_. “All this takes away from the artist's original design intent on the character and its shape language, with what pretty much functions on a surface level as a 'Snapchat filter.’”