Generating Long Videos of Dynamic Scenes

Dataset Analysis

A video generation model that redesigns temporal latent representation and uses two-phase training to generate long videos with consistent object motion, camera viewpoint, and new content.