Last month, Nvidia revealed its latest version of DLSS, which has received negative reviews from players and developers alike. The criticism stems from the tool appearing to change the art direction and style of the games Nvidia used to showcase the technology, including Resident Evil Requiem and Starfield.
Since the DLSS 5 announcement, Nvidia CEO Jensen Huang has responded to public outcry a handful of times, saying gamers are completely wrong about their attitude toward the tool. In his latest comments, Huang, unsurprisingly, once again defended DLSS 5, saying it's just another way for developers to improve visual fidelity. He took it a step further by saying that he sees all AI-generated content as “beautiful.”
However, developers like Quin Henshaw, Unity development engine expert and Unity instructor at the Vancouver Film School, see the latest version of DLSS as a step in the wrong direction. From his point of view, DLSS 5 will only have a negative impact on game development and art direction as a whole.
What is DLSS?
Nvidia's deep learning super sampling (DLSS) is an AI-powered graphics technology designed to both increase visual fidelity and improve performance. It uses AI to render games at a lower resolution and then upscale the image to make it look better. It's similar to Sony's PSSR, which also leverages AI to go pixel-by-pixel to exclusive resolution.
DLSS uses a type of AI called a deep neural network (hence the “deep learning” part of its name) to enhance images. DLSS was trained on supercomputers to recognize high-quality images and learn to reconstruct which lower-quality images were missing, therefore upscaling the image.
It's not something developers use to build games, but it's something that runs with the game to improve its image and performance. Developers can set it to varying degrees, but DLSS must be implemented and enabled by developers to work.
DLSS improves performance by reducing the amount of work a computer's GPU must do. If a game is rendered in 4k, it means that the GPU is processing millions of pixels. With DLSS, the game can be rendered in 1440p or even 1080p, and then use AI to upscale it to 4k.
The previous versions of DLSS have not been problematic. In fact, image upscaling is becoming more common and is generally seen by developers as a positive technological advance in terms of game fidelity. “All that upscaling technology is pretty clever, and as much as [NVIDIA is] leading with it, pretty much every software developer has their own version of it now,” says Henshaw.
AMD, for example, offers FidelityFX Super Resolution (FSR) as an upscaling tool. However, it doesn't require dedicated AI hardware like Tensor Cores, which is what DLSS needs to run.
“I've been pretty much on board with a lot of the technology that Nvidia has created over the last few years,” Henshaw said. He has used previous versions of DLSS to create technical demos in the past. Generative AI was first introduced in DLSS in 2022 with DLSS 3. It marks the point where Nvidia moved from AI-assisted rendering to AI-generated frames.
DLSS 3 used generative AI for frame generation. The GPU creates a frame to go between frames rendered by the game's engine to improve image quality and performance. DLSS 3.5 took it a step further by allowing light generation via Ray Reconstruction. DLSS 4 improved the technology with improvements to frame generation and greater integration with Ray Tracing technology.
However, DLSS 5 does more than just upscale an image via generative AI to make a game run smoother or look better. It starts to more aggressively affect how the final image looks, potentially changing aspects of how a scene looks. That's what developers and players have trouble with.
“It's not something that I could see any serious development studio, whether at the indie level or triple-A level, really getting into, unless there's a lot of pressure from publishers to cut costs,” Henshaw said.
In a video by YouTuber Daniel Owens, Nvidia's Jacob Freeman provided some more details on how DLSS 5 works. It doesn't really change anything at the game engine level, but it essentially takes a 2D image and runs it through its generative AI technology to adjust the image. Artists can control things like color gradients and filter intensity, but they don't seem to have control over the final result. Freeman also stated that this is an early preview of the technology. DLSS 5 is expected to be released sometime this fall, so it stands to reason that the full scope of developer controls will be clarified around that time.
Developers claim that DLSS 5 takes away their autonomy
DLSS 5 is more aggressive with its AI generation than previous iterations of the tool. It has been shown to enhance lighting, materials and fine details, from enhancing an image to potentially influencing how these elements are displayed. It uses what it has been trained on to improve the appearance of an image to make it appear higher quality.
That's why Grace's face looks so different from her original image – because that's how DLSS 5 thinks it should look. The end result has been panned by players and developers as “AI Slop”, and is similar to a filter you might see on Instagram. “I feel like anyone who's at least a little bit familiar with the internet and gamers would have known that it would have just instantly been thrown all over the place. So it was kind of shocking to me,” Hendshaw said.
It was also where Nvidia CEO Jensen Huang first addressed public reaction to the technology. “Well, first of all, they're dead wrong,” he said in an interview with Tom's Hardware. Huang later softened his comments in an interview with Lex Fridman, rephrased as an understanding of the criticism, but still stated that he sees all generative AI as “beautiful”.
Huang wasn't the only executive taken with the technology's potential. “When Nvidia showed us DLSS 5, and we put it to work in Starfield, it was amazing how it brought it to life. We've been playing it. We can't wait for you all to too,” said Todd Howard, executive producer and game director at Bethesda, during DLSS 5's initial reveal. Starfield was recently used in a DLSS 5 tech demo, with 12 minutes of the technology in action.
It goes back to something pointed out by Mat Piscatella, a video game industry analyst and managing director of gaming at Circana. It may seem obvious, but he said, “CEOs will generally be optimistic about a technology that can save them a lot of money.” Which is a sentiment that has artists like Henshaw worried about the potential ramifications of Nvidia's latest technology.
“The difference with DLSS 5 is that it pushes into artistic quality and takes control away from developers,” Henshaw said. “I see it as a big, big deal. If studios start implementing it, they're probably going to cut a huge number of artists.”
Henshaw worries that managers will see this tool as an opportunity to reduce labor and costs. Even if the final product does not fare well for consumers, the cost saved on paying artists may balance the loss in the eyes of management. He also said that consumers in general are already quite fed up with how much AI-generated content is out there and are likely to respond with their wallets.
In Huang's latest interview, he positioned DLSS 5 as another tool to be used by developers to improve the visual fidelity of their games. But Henshaw argues that it will only move the artistic needle in a more homogenous direction, which is a common criticism of generative AI.
Generative AI in video games is generally frowned upon by society. Recent controversies include Crimson Desert developer Pearl Abyss uses gen AI assets in its finished game. The developer took to Twitter to apologize, but it's only the latest instance that drew backlash from players. Earlier controversies arose with Baldur's Gate 3 developers Larian Studios and Clair Obscur: Expedition 33 developer Sandfall Interactive.
The problem with developer adoption of DLSS 5 underscores a growing concern in game development as a whole. As AI tools are increasingly integrated into game developers, developers have little say in using them. While most developers don't seem to mind AI being used as a tool in workflows like programming or early conception stages, they generally oppose generative AI tools used in the creative side of things, such as voice acting, art, mission design, and asset generation.
“With the industry the way it is right now, [developers have] very little pushback. Maybe in smaller studios, but in bigger studios there will be zero pushback. There is very little advocacy and power on the part of developers when it comes to major publishers and studios,” said Henshaw.
Ultimately, DLSS 5 highlights a growing divide in how AI is viewed across the industry. While companies like Nvidia and potentially Bethesda see it as a way to push visual fidelity forward and streamline development, many artists and developers worry about what might be lost in the process. Whether DLSS 5 is widely adopted or faces resistance will likely depend on how much control managers are willing to give up, and how players react to AI having a bigger hand in shaping what games actually look like.
