Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

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

Source

arXiv

arxiv.org/abs/2607.00918v1

Read the source

Source

Plain-Language Brief

A new framework called MAGNET is introduced for generating long-form narratives using multiple agent characters that maintain consistency and coherence in their actions based on a shared world state and evolving story goals. This framework reduces annotations and hallucinations compared to single-model baselines and IBSEN, according to evaluations.

What the Document Says

  • A new narrative generation and verification framework called MAGNET is introduced.
  • MAGNET uses multiple agent characters grounded in personas that propose actions based on a shared world state and evolving story goals.
  • ATLAS is a graph-based pipeline used for comparing scene-level world representations across a generated story to detect hallucinations.
  • The study evaluates MAGNET using an LLM editor, pairwise rubric scoring, and ATLAS.
  • Compared to single-model prompting and IBSEN, MAGNET reduces annotations by 41% and 34%, respectively, and reduces hallucinations by 50% and 45%, respectively, in long-form narratives of 100 pages.
  • Pairwise rubric evaluation shows similar results.

Key Concepts Mentioned

  • narrative generation
  • multi-agent storytelling
  • world state tracking
  • evolving story goals
  • hallucinations detection

Claims Made by the Document

  • MAGNET produces coherent narratives compared to single-model prompting and IBSEN.
  • Long-form narratives can emerge from explicit world-state tracking and goal-driven multi-agent generation.

Evidence Used

  • Reduced annotations and hallucinations in long-form narratives of 100 pages compared to single-model baselines and IBSEN.
  • Pairwise rubric evaluation showing similar results.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

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.