Scaling Long-Form Story Generation via Narrative State Tracking

2026-09-28

AI Summary

This study addresses the challenge of maintaining narrative consistency when large language models generate long-form novels, noting that existing approaches are largely confined to texts under 10,000 words. To overcome this limitation, this work proposes NstAgent, a novel training-free, agent-based narrative state tracking framework. By dynamically maintaining structured narrative states—such as characters and events—the method enables effective consistency control throughout extended text generation. Experimental evaluations across scales ranging from 10,000 to 100,000 words demonstrate that the proposed approach significantly enhances both narrative coherence and overall writing quality. Consequently, this work effectively transcends current scalability bottlenecks in long-form novel generation and validates the feasibility of extending automated storytelling to full-length novels.

Abstract

LLMs have demonstrated strong capabilities in creative writing. However, scaling them to full-length novels remains challenging, as maintaining narrative consistency becomes increasingly difficult. Existing story-generation methods typically focus on stories of up to about ten thousand words, leaving their ability to scale to full-length novels underexplored. In this work, we introduce Narrative State Tracking Agent (NstAgent), a training-free agentic framework that allows LLMs to track a structured narrative state including characters, past events and future requirements. We extend an existing benchmark to compare narrative consistency across lengths, and use it together with a writing-quality benchmark to systematically evaluate stories ranging from 10K to 100K words. We show that NstAgent achieves better narrative consistency and writing quality as stories grow longer, and neither of them degrades noticeably as length increases, suggesting that it provides an effective approach to scaling story generation toward full-length novels.

Problem

• Long-form story generation
• Narrative consistency
• Large language models
• Full-length novels

Innovation

• Narrative State Tracking
• Training-free Agent
• Long-Form Story Generation
• Narrative Consistency
• Large Language Models