Nvidia CEO Jensen Huang has clarified that he’s additionally not a fan of AI slop, however that DLSS 5 isn’t aiming to show video games into it. As a substitute, it is a instrument that builders can, or can not, select to make use of for growth.
Showing on Lex Fridman’s newest podcast episode, Huang was requested about DLSS 5 and the criticism of it, per week after Nvidia introduced the brand new AI-powered instrument at GTC. Huang acknowledged that he now understood the place the criticism stemmed from, only a few days after proclaiming all critics of DLSS 5 as “utterly fallacious.”
“I believe their perspective is sensible, and I can see the place they’re coming from, as a result of I do not love AI slop myself. You realize, the entire AI-generated content material more and more seems to be related, they usually’re all stunning, and I can…so I can…I am empathetic to what they’re considering,” Huang acknowledged.
Talking on the potential for DLSS 5 to show video games into AI slop, Huang reiterated that the instrument was absolutely within the fingers of builders and below their management.
“DLSS 5 is 3D-conditioned, it is 3D-guided. It is floor reality construction data-guided. The artist determines the geometry, we’re utterly truthful [to the geometry in every frame],” he continued. “It is conditioned by the textures, the artistry of the artist. And so each single body, it enhances however would not change.”
This aligns with Nvidia’s messaging around DLSS 5 since its announcement, during which it explicitly acknowledged that the expertise is ready to improve options such because the sheen on a cloth, the interplay of sunshine on hair, and extra with out altering the construction and semantics of the unique scene. It additionally acknowledged within the announcement that builders can have management over completely different facet of the implementation, equivalent to depth, colour grading, masking, and extra.
This, nonetheless, stands in distinction to feedback made by one other Nvidia worker, who additional detailed how DLSS 5 works with YouTube creator Daniel Owen. Talking to Jacob Freeman, a GeForce Evangelist at Nvidia, Owen requested bluntly whether or not DLSS 5 was taking a static 2D rendered body from a recreation and utilizing that as the only real enter for the mannequin to reinforce. “Sure, DLSS 5 takes a 2D body plus movement vectors as enter,” Freeman replied through e-mail.
Owen then adopted up by asking if DLSS 5 has any entry to the underlying 3D geometry knowledge current in a recreation’s engine, or whether or not DLSS is just like taking a screenshot and asking a generative AI mannequin to reinforce it. Freeman’s reply right here aligned with Nvidia’s earlier feedback, stating that the mannequin is skilled to grasp advanced scenes, however nonetheless concluded by saying that that is all accomplished by “analyzing a single body,” inferring that the mannequin has no fast info supplied to it by the sport it’s applied in exterior of the presently rendered body, and the one instantly after it.
This might clarify why, as Owen mentions all through his video, character fashions have particulars, equivalent to further hair and make-up, that aren’t in any respect current on the unique 3D fashions, with DLSS 5 as an alternative inferring these enhancements from the idea of a single body alone. That makes it sound far much less refined than Huang is describing on this interview, however would possibly clarify why he thinks the instrument is likely to be helpful for artists to reinforce their video games with wholly completely different artwork kinds to what was initially meant.
“…sooner or later you would immediate it, ‘I need it to be a toon shader.’ I need it to seem like this, kinda, , so that you may give it an instance. And it might generate within the fashion of that. All per the artistry, , the fashion, the intent of the artist. All of that’s accomplished for the artist, in order that they will create one thing that’s extra stunning, however nonetheless within the fashion that they need,” Huang concluded.
In October 2025, the Content material Abroad Distribution Affiliation (CODA) despatched an open letter to OpenAI requesting the corporate to stop coaching its fashions on copyrighted content material it represented, within the wake of the overwhelming inflow of photographs and movies generated utilizing the toon artwork fashion from Studio Ghibli movies. OpenAI had beforehand allowed customers to immediate its newest Sora 2 mannequin to use the shader to supplied photographs and movies.