From NVIDIA: 2025-06-11 07:00:00

NVIDIA Research introduces DiffusionRenderer, an AI light switch for videos that can alter lighting in real or AI-generated videos, aiding creative industries and physical AI development. The tool combines inverse and forward rendering processes, presenting a powerful framework for video lighting control and editing. Over 60 NVIDIA papers, including DiffusionRenderer, accepted at CVPR conference.

DiffusionRenderer revolutionizes de-lighting and relighting scenes from 2D video data, offering realistic changes in lighting conditions without the need for 3D geometry data. AV developers can utilize this tool to generate diverse lighting scenarios for training autonomous vehicle models. It enhances creative content creation processes by allowing early exploration of various lighting options.

NVIDIA Research integrates DiffusionRenderer with Cosmos Predict-1, enhancing de-lighting and relighting quality with a scaling effect. Cosmos Predict is part of NVIDIA Cosmos, a platform accelerating synthetic data generation for physical AI development. NVIDIA researchers at CVPR present groundbreaking papers in automotive, healthcare, and robotics, with three nominated for Best Paper Award.

NVIDIA secures Autonomous Grand Challenge win at CVPR for the second year in a row, showcasing leadership in end-to-end autonomous systems development. Explore NVIDIA Research’s global team and their innovative work in AI, computer graphics, computer vision, and self-driving cars. Dive into the latest research papers presented at CVPR and watch the NVIDIA GTC Paris keynote by CEO Jensen Huang.



Read more at NVIDIA: NVIDIA Research Unveils Neural Rendering Model for Physical AI