Lighting-grounded Video Generation with Renderer-based Agent Reasoning

Dataset Analysis

LiVER is a diffusion-based framework for scene-controllable video generation using explicit 3D properties, dense annotations, and an agent for user instructions, achieving disentangled control.