Multi-Shot Video Generation
Overview
The field studies how to generate videos made of multiple shots that remain coherent in story, appearance, motion, and cinematic style. Common approaches use scripts, storyboards, keyframes, and camera plans to structure shots, alongside diffusion or autoregressive models conditioned on entity memories and retrieved visual references. Agentic systems also plan, generate, inspect, and refine sequences, while benchmarks increasingly assess identity continuity, narrative logic, shot transitions, camera control, editing, and audio-visual alignment. A central challenge is sustaining causal and visual consistency over long sequences without sacrificing shot quality, controllability, or generation efficiency.
Latest papers
70 papers13 papers in the last four weeks, up 44% on the four weeks before. 0.1% of all new papers.