On March 16th, Nvidia officially presented DLSS 5 and promptly received a lot of criticism. With the help of AI, games will be supported by “photorealistic lighting and materials”. However, the Internet sees more AI slop and uniformity and less improvement or prettification. Nvidia CEO Jensen Huang then explained that we had misunderstood the concept. However, a community and marketing manager from Nvidia now contradicts this statement.
AI filter or deep engine tool?
The idea of DLSS 5 certainly wasn’t as well received as Nvidia had hoped. The internet was full of criticism and malice, because slapping an ugly AI filter on games was not an innovation or improvement. Martin also writes in his article that the whole thing certainly has potential as a tool if used correctly. However, there is also the risk of turning all games into a generic uniformity that is ironed out in the same style. Individuality and your own styles would be lost.
Shortly afterwards, Nvidia CEO Jensen Huang defended himself against this representation and said that everyone who criticized it was completely wrong. “The reason for this is that, as I explained very carefully, DLSS 5 merges the controllability of the geometry and textures and everything about the game with generative AI,” he explains.
Basically, this means that, according to him, DLSS 5 doesn’t just work superficially, taking an image, analyzing it and then putting a filter over it, but that developers have full control. DLSS 5 interacts deeply with the game and the technology and not just with the individual frames.
Only the finished image is interpreted
Now, however, Jacob Freeman has contradicted exactly this representation in an interview. In response to technical YouTuber Daniel Owens’ questions, he basically said that DLSS 5 actually only processes individual frames and motion data instead of going all the way down to the geometry level.
Watch on YouTube
“Yes, DLSS 5 uses a 2D image and motion vectors as input,” explains Freeman. “DLSS 5 has been consistently trained to recognize complex scene features such as characters, hair, fabrics and translucent skin, as well as ambient lighting conditions such as front or backlight or overcast skies – all by analyzing a single image.”
So instead of sitting “deeper” in the engine and interoperating with the systems, as Huang portrays, Freeman’s version seems much closer to the gaming community’s fears.
Are Freeman’s statements trustworthy?
While his depiction isn’t exactly anti-Huang, it doesn’t exactly support the picture Nvidia is trying to paint either. With his years of technical expertise, Freeman can certainly be trusted. He has been a so-called GeForce Evangelist at Nvidia since 2023, meaning he works in technology marketing, community management and explains the new products. He had previously spent many years in the PC and technology industry.
Freeman is not a blank slate, so his statements seem quite realistic. However, this is not a direct confirmation. While Freeman’s statement is not the same as the CEO’s, it does not openly contradict him. However, we can assume that we are really dealing with a filter technology rather than a deeply rooted engine tool.
