Google Research Introduces an AI Video Co-Director: 4 Agentic Frameworks for Coherent, Minutes-Long Video Generation

Google Research has introduced an AI system with four frameworks designed to generate coherent videos that are several minutes long rather than just short clips. The challenge it addresses is that existing multi-shot video pipelines fail because characters and scenery shift between shots and errors in early footage corrupt all later shots. The system works as an orchestration layer on top of existing models, using persistent visual memory to track characters and locations throughout a story, a bandit search approach for creative planning, and a prompt refinement loop that critiques and improves text descriptions. Code for some components is publicly available, though the full pipeline is not yet a Google product.