NVIDIA's DiffusionRenderer:AI Rendering for Robotics, Image Editing,and Autonomous Vehicles

​Our guest, Zan Gojcic, a Senior Research Manager at NVIDIA, will present to the BuzzRobot community DiffusionRenderer, which was recently introduced at CVPR.

​NVIDIA’s DiffusionRenderer is a neural rendering framework that unifies inverse and forward rendering to enable advanced video editing, content creation, and synthetic data generation for autonomous vehicles and robotics.

​Unlike traditional physically-based rendering (PBR), which requires precise 3D geometry, material properties, and lighting conditions, DiffusionRenderer leverages video diffusion models to approximate light behavior directly from real-world videos. It can transform day into night, simulate weather changes, and soften harsh lighting, offering tools for video relighting, material editing, and realistic object insertion.

​By accurately estimating G-buffers from video input, it supports both creative applications and physical AI development (autonomous vehicles, robotics), outperforming state-of-the-art methods.

​Read the paper and the blog post.
Join the BuzzRobot Slack to stay connected with the community.