Overview
- At SIGGRAPH on Tuesday, Nvidia showed that developers will get two global sliders — Structural Intensity and Tone Intensity — to control how strongly DLSS 5's generative pass alters high‑frequency detail and overall lighting and color.
- Nvidia demonstrated three separately trained models labeled A, B and C that studios can swap in real time for different scenes or levels of change without adding latency.
- Developers can mask individual objects or characters and set independent Structure and Tone values per mask so artists can exclude or limit DLSS 5 on faces while enhancing environments.
- Nvidia says DLSS 5 starts from the conventional rendered frame and then applies a learned generative stage anchored by internal buffers and motion vectors that process frames causally on a single GPU, but the company has not published performance or VRAM minimums.
- The company expects a Q3/fall 2026 rollout and says it is refining controls with partners, leaving open whether the exposed sliders and masks will satisfy critics who objected to earlier demos that changed character appearance.