Nvidia's DLSS 5 appears to rely on single-frame AI inference

After attempts to get answers from Nvidia, techtuber Daniel Owen obtained comments from GeForce evangelist Jacob Freeman, who says, "DLSS 5 takes a 2D frame plus motion vectors as input." Demos shown at GTC ran on two RTX 5090 cards—one to render the game and another $4,000 GPU to run the DLSS 5 compute path—yet the preview iteration appears to be operating on a static 2D image and motion vectors, effectively applying an AI filter in real time.

That design leaves the model without direct awareness of 3D geometry, depth, or any lighting information beyond what’s visible in the frame. Freeman says the model is trained to infer "complex scene semantics such as characters, hair, fabric and translucent skin, along with environmental lighting conditions like front-lit, back-lit or overcast—all by analysing a single frame," and that "Materials are inferred from the rendered frame." It does not have hooks into the game engine to receive ground-truth data.

nvidia, dlss 5, ai inference, single-frame, motion vectors, rtx 5090, geforce, gtc, scene semantics, game engine