Nvidia DLSS 5: Artists Gain Control Over AI Graphics

Nvidia has revealed significant updates to its DLSS 5 technology, aiming to quell widespread concerns raised after its initial announcement. At a recent SIGGRAPH keynote, the company emphasized enhanced control for artists, a key point of contention for gamers and developers.

Artists will now have granular control over DLSS 5's graphical output. This includes the ability to mask individual characters, faces, props, or background elements, allowing for tailored AI processing. Developers can utilize Structural and Tone Intensity sliders and select from three distinct AI models to shape the final visuals, ensuring the AI enhances rather than overwrites artistic direction.

Nvidia clarified that DLSS 5 does not generate game scenes from scratch but strictly uses internal game engine buffers to maintain the original intent of developers. Traditional rasterization and ray tracing still form the frame's foundation, with DLSS 5 acting as a third AI rendering category that enriches complex lighting elements without hallucinating or rewriting visuals.

Addressing latency, Nvidia confirmed DLSS 5 processes frames using native motion vectors from the game engine, avoiding multi-frame batch processing. This approach is designed to eliminate input lag and temporal shimmering artifacts, crucial for fast-paced gaming experiences.

Finally, the requirement for dual high-end GPUs has been eliminated. Nvidia has distilled the generative model into a smaller, single-step diffusion model that can run efficiently on a single RTX GPU, making the technology more accessible to a wider audience. The company is committed to delivering richer game graphics without the previously criticized AI aesthetic.