graphical

Nvidia presented DLSS 5 at GTC 2026 as the greatest graphical advance that video games have experienced since the ray tracing. The reaction has been almost unanimous: the gaming community and industry professionals themselves have described it as a “slop AI filter.” The rejection has been so frontal and almost unanimous that Nvidia has had to come out to clarify how the technology works and what control developers really have over these visual improvements. What is DLSS 5. DLSS technology was born in 2019 as an intelligent upscaling system: the GPU renders at lower resolution and the AI ​​reconstructs each frame up to 4K with minimal quality penalty. With each iteration (DLSS 3, 3.5, 4, 4.5) the goal remained the same, but DLSS 5 breaks that logic. According to Nvidia’s own announcementwe are looking at a real-time neural rendering model that analyzes the color and motion vectors of each frame and generates lighting and photorealistic-looking materials on them. The system recognizes the semantics of the scene (skin, hair, fabrics, metals) and applies its own interpretation of how those elements should look under real physical lighting. Jensen Huang defined it with a phrase that well summarizes the ambition of this new iteration: “Twenty-five years after Nvidia invented the shader programmable, we are reinventing computer graphics.” Digital Foundrywhich had access to the technology before the announcement (and which has been heavily criticized for its glowing coverage), called it “the most amazing I’ve seen in my time at Digital Foundry” and pointed to genuinely notable improvements in environments from ‘Assassin’s Creed Shadows’ or ‘Oblivion Remastered’. The faces, Juan, the faces. The problem is that the official demo video included sequences from ‘Resident Evil Requiem’, ‘Starfield’, ‘Hogwarts Legacy’ and ‘EA Sports FC’, and in all of them the system visibly altered the characters’ faces. The protagonist of ‘Requiem’, Grace Ashcroft, has been the most widespread example: more pronounced cheekbones, fuller lips and uniform skin tone. According to Kotakuthe effect seems to apply a TikTok beauty filter on characters with an artistic intention other than physical attractiveness, as is the case with Ashcroft. Another example is that of the ‘Starfield’ characters, which are not very detailed in themselves, and which gain facial resolution but lose all aesthetic coherence with the original design. In ‘Hogwarts Legacy‘, an old woman with gently modeled wrinkles begins to show off a deeply cracked face completely alien to what was seen in the game. Therefore, the dreaded term ‘AI slop‘ appeared on social networks in a matter of minutes. He Nvidia GeForce on X announcement post was buried by negative replies, which accumulated favs and RTs in much greater quantity than the original post. Also the comments of the Digital Foundry video They were almost unanimously negative. The answer. Given the volume of criticism, Nvidia published a statement on YouTube clarifying how the system works. According to the companydevelopers have complete artistic control over DLSS 5: they can adjust the intensity of the effect, the color grading and mask specific areas where they do not want the AI ​​to act (the company calls it “controllability”). The company also clarified that the technology is not a filter applied on top of the image, but rather takes the color and motion vectors of the game to generate its output, “anchored to the source 3D content.” Bethesda, one of the most active studios in the initial support (Todd Howard had appeared in the presentation video praising the results in ‘Starfield’) posted hours later a more nuanced response on the studio’s official account. There they stated that “our art teams will adjust the lighting and final effect to look the way we consider best for each game. Everything will be under the control of our artists and will be completely optional for players.” Two ways of looking at it. The disparity in reactions reflects two legitimate ways of evaluating the same technology. What a good part of the community and numerous media outlets have criticized is that the modifications make the characters more realistic but different from how they were designed by the game’s art team. For example, concept artist Jeff Talbot said that: “In each shot the artistic direction was removed to add meaningless ‘details’ (…) This is a garbage AI filter.” Poor optimization. a few weeks ago There began to be talk that the proliferation of tools of upscaling and AI has reduced the pressure on studios to optimize their games: when DLSS or FSR can more than compensate for performance issues, the incentives to polish the native engine disappear. There is already someone he says it bluntly: Some studios design their games from the beginning assuming that the upscaling It will fix what’s broken, rather than using it as a further improvement on an already solid foundation. With DLSS 5 that takes a qualitative leap, and the risk is not only aesthetic: it is work-related and creative. And then there’s an additional detail: the GTC demo required two GeForce RTX 5090s running in parallel (one to render the game, another to run the DLSS 5 neural model). Nvidia claims that the final launch, scheduled for fall 2026, will work with a single card, but the magnitude of the hardware raised questions about the actual requirements. If studios start designing with DLSS 5 as a safety net, what version of the game will the player without that GPU receive? Real video games. There is something that Nvidia seems to have not taken into account: people like video games because they look like video games. Imperfection has a human touch that is part of the product’s identity. Grace Ashcroft works as a character in ‘Requiem’ precisely because her appearance reflects exhaustion and vulnerability. DLSS 5’s AI makes it something that has been described as the result of applying Nvidia’s system to a character whose aesthetic is not designed to be photorealistic. The problem isn’t just that the result is aesthetically questionable: it’s that the entire premise is wrong. Nvidia assumes that “more … Read more