Research & Innovation
Google's Fix for Long-Form AI Video Is Four Agents Arguing About Continuity
Google Research published a multi-agent framework for coherent long-form video generation on September 24: an AI video co-director selecting creative configurations via multi-armed bandits, CANVAS holding persistent visual memory of characters and locations, A²RD generating long sequences through retrieve-synthesise-refine-update loops, and VQQA refining prompts from vision-language critiques.