NVIDIA Showcases DLSS 5 Improvements with Real-Time Model Switching at SIGGRAPH

At the SIGGRAPH conference, NVIDIA demonstrated new developer controls for DLSS 5, allowing real-time switching between three distinct AI models (Model A, B, and C) to adjust rendering styles. Developers can now fine-tune visual outputs using "Structure Intensity" and "Tone Intensity" sliders to customize details and lighting. These updates address previous criticisms of DLSS 5 by giving developers granular control over AI-generated graphics, ensuring the technology respects the game's original artistic vision. This level of customization helps bridge the gap between automated AI upscaling and intentional art direction in modern game development. The "Structure" slider controls high-frequency details like subsurface scattering and reflections, while the "Tone" slider manages low-frequency elements like broad lighting and color. NVIDIA also highlighted challenges in achieving sub-16ms latency for 4K rendering at over 60 FPS while maintaining temporal consistency using motion vectors.