DramaGAT: A Hierarchical Multi-Agent Collaborative Framework for Long-Form Script Generation

Despite the widespread application of large language models (LLMs) in creative writing tasks, challenges remain in maintaining coherence and consistency during longform script generation. In this work, we propose DramaGAT (Drama Generation AgenTs), a hierarchical multi-agent collaborative framework which effectively increases the length of generated scripts while ensuring good coherence and consistency. DramaGAT simulates the workflow of professional screenwriters by decomposing the scriptwriting process into multiple key stages, each handled by specialized agents. To evaluate this framework, we conduct a systematic analysis of the generated scripts based on a four-dimensional scoring standard, combining LLM-based automatic evaluation with human assessment. The results show that our method significantly extends the length of the generated scripts while maintaining coherence, consistency, and content quality, demonstrating its practical value for long-form script generation tasks.