Long-form video generation struggles with narrative planning and visual consistency across scenes. ViMax is a multi-agent framework that decomposes the pipeline into specialized agents for screenwriting, shot planning, character styling, video generation, and VLM-based quality control. These agents collaborate to maintain storytelling integrity and visual coherence.
The framework uses a hierarchical narrative engine with retrieval-augmented generation to keep local decisions grounded in global story context. For visual consistency, ViMax employs graph-based dependency tracking to preserve character and object identity across shots, and generates transition videos to maintain spatial coherence between camera angles. This agentic approach addresses two fundamental challenges in long-form video: multi-shot visual consistency and narrative planning over dozens of shots.