Google researchers demonstrate ten-minute AI-generated video

Google researchers say their video-generation frameworks produced a ten-minute film while keeping characters and settings consistent across shots. The work combines planning, visual memory and repeated review of generated material to address the drift that can appear as a video grows longer. One component tracks characters, locations and objects so later shots can reuse their visual details. Another generates video in segments, drawing on earlier segments when a scene or character returns. A review system asks visual questions about the result and revises prompts for another attempt. Google reports gains on several research benchmarks; it describes the systems as research frameworks and says it is exploring how to incorporate human creative workflows.