Overview
- Nvidia clarified at SIGGRAPH on Tuesday, July 21, 2026, that DLSS 5 will give developers two global intensity sliders named Structure and Tone to control how strongly the model alters a frame.
- The Structure slider changes high-frequency details such as ambient occlusion, reflections, contact shadows and subsurface scattering while the Tone slider adjusts lower-frequency elements like lighting, color and overall mood.
- Studios can mask individual objects or characters and apply independent Structure and Tone settings, and they can swap among three trained models (A, B and C) to choose different visual interpretations per scene.
- Nvidia describes DLSS 5 as a one-step pixel-space diffusion transformer that starts from a conventionally rendered 2D frame plus motion vectors and uses internal buffers (albedo, normals, lighting) to anchor results to artists' intent.
- Nvidia says the model is designed to run on a single GPU and that more developer controls are in development, but the company has not yet released concrete performance or VRAM figures so community testing after the planned fall launch will determine real-world impact.