From what Nvidia’s showing (and what Huang emphasized in that Tom’s Hardware Q&A at GTC 2026), DLSS 5 is pitched as “neural rendering” or “content-control generative AI.” It uses the game’s existing 3D data (motion vectors, scene semantics like hair/skin/fabric, lighting conditions) as grounded input, then the AI infuses photoreal lighting, materials, and enhancements while staying consistent frame-to-frame. Devs get tools to fine-tune intensity, color grading, masking, etc., so they can dial in how much “photorealism” gets applied without losing their artistic intent