From Personas to Plot: Character-Grounded Multi-Agent Story Generation for Long-Form Narratives

2026-07-01

AI Summary

This work addresses the challenges of narrative inconsistency and hallucination commonly encountered when large language models generate long-form stories. The authors propose MAGNET-ATLAS, a unified framework that integrates goal-driven multi-agent story generation with a graph-based verification mechanism. In this framework, MAGNET enables character agents to collaboratively craft narratives through shared world states and dynamic objectives, while ATLAS employs graph neural networks to automatically detect scene-level hallucinations. Evaluated on hundred-page-scale stories, the method significantly outperforms both single-model baselines and IBSEN, reducing annotated errors by 41% and 34%, and hallucinations by 50% and 45%, respectively. Human evaluations further confirm a marked improvement in narrative coherence.

Abstract

Although large language models (LLMs) have demonstrated impressive creative fiction generation, they struggle to maintain narrative consistency and coherent plot lines in long-form stories. In this work, we introduce a unified framework for long-form narrative generation and verification. MAGNET, a multi-agent goal-driven narrative engine for storytelling, generates stories with persona-grounded character agents that propose actions based on a shared world state and evolving story goals, while ATLAS is a graph-based pipeline that compares scene-level world representations across a generated story to detect hallucinations. By evaluating MAGNET using an LLM editor, pairwise rubric scoring, and ATLAS, we show that our framework produces coherent narratives compared to single-model prompting and IBSEN. At 100 pages, MAGNET reduced annotations and hallucinations by 41 and 50%, respectively, compared to the single model baseline and by 34 and 45%, respectively, compared to IBSEN, with pairwise rubric evaluation showing similar results. These results suggest that long-form narratives can emerge from explicit world-state tracking and goal-driven multi-agent generation, providing a foundation for controllable and structurally coherent long-form narrative generation.

Problem

• long-form narrative generation
• narrative consistency
• coherent plot
• hallucination
• storytelling

Innovation

• multi-agent storytelling
• persona-grounded generation
• world-state tracking
• hallucination detection
• long-form narrative