Hollywood Town: Long-Video Generation via Cross-Modal Multi-Agent Orchestration

TL;DR

This work tackles the challenge of minute-scale long-video generation from text by introducing OmniAgent, a hierarchical, graph-based multi-agent system that mirrors film production workflows. It couples LLM planning with multimodal back-ends and a novel context-engineering layer, enabling on-demand cross-agent collaboration through hypergraphs and bounded cyclic execution for graph-level reflection within a finite retry budget. The key contributions are the OmniAgent framework, hypergraph-based context retrieval, and controlled cyclic execution, supported by comprehensive human evaluations showing improved structure, audiovisual expressivity, and engagement compared to baselines. The findings demonstrate that orchestration quality—particularly hierarchy and context-driven reflection—drives cinematic outcomes, suggesting practical pathways for scalable, robust multi-agent systems in creative tasks. This approach advances long-video generation by focusing on the orchestration layer, offering a principled way to balance context, memory, and iterative refinement in complex, cross-modal pipelines.

Abstract

Recent advancements in multi-agent systems have demonstrated significant potential for enhancing creative task performance, such as long video generation. This study introduces three innovations to improve multi-agent collaboration. First, we propose OmniAgent, a hierarchical, graph-based multi-agent framework for long video generation that leverages a film-production-inspired architecture to enable modular specialization and scalable inter-agent collaboration. Second, inspired by context engineering, we propose hypergraph nodes that enable temporary group discussions among agents lacking sufficient context, reducing individual memory requirements while ensuring adequate contextual information. Third, we transition from directed acyclic graphs (DAGs) to directed cyclic graphs with limited retries, allowing agents to reflect and refine outputs iteratively, thereby improving earlier stages through feedback from subsequent nodes. These contributions lay the groundwork for developing more robust multi-agent systems in creative tasks.