EUVIMEDEUVIMED entdecken
Crossref

GRAIL-VN: A Graph-Grounded Runtime-Verifiable Agentic Framework for Reliable Branching Visual Novel Script Generation

Chuanru Wei, Kaili Chen

Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · 08.11.2026 · 22 (1) · 424-433

DOI: 10.1609/aiide.v22i1.43600 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, online erschienen am 08.11.2026. Verfasst von Chuanru Wei und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The core challenge in visual novel script generation lies not only in generating coherent narrative content, but also in continuously maintaining the consistency of branching structures, character states, and conditional logic as the story evolves. This paper proposes GRAIL-VN, a Graph-Grounded Runtime-Verifiable Agentic Framework for Reliable Branching Visual Novel Script Generation, which models visual novel script generation as an iterative update process based on narrative state. The framework maintains persistent narrative state through a Narrative World Graph (NWG), performs state retrieval, local generation, and state commitment through a graph-backed iterative runtime (GIR), and applies multi-layer runtime verification before updates are written back through Plan-Graph-Constrain-Realize (PGCR); Lookahead Branch Planning (LBP) and Just-In-Time Casting (JITC) further support lookahead branch decision-making and dynamic expansion of the narrative world. We conduct experiments on 36 narrative briefs covering different genres and complexity levels, and perform an integrated analysis combining automated structural evaluation, LLM pairwise narrative quality evaluation, and goal achievement evaluation. The results show that the value of the persistent graph mainly comes from its continuous participation as a runtime state interface, rather than from one-time reading during the generation stage; PGCR can effectively detect and repair inconsistent updates, thereby improving generation reliability, while LBP and JITC further enhance branch planning and dynamic expansion capabilities. Overall, GRAIL-VN provides a unified runtime framework for reliable branching visual novel script generation.

Zitieren

APA

Wei, C., & Chen, K. (2026). GRAIL-VN: A Graph-Grounded Runtime-Verifiable Agentic Framework for Reliable Branching Visual Novel Script Generation. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 22(1), 424-433. https://doi.org/10.1609/aiide.v22i1.43600

Vancouver

Wei C, Chen K. GRAIL-VN: A Graph-Grounded Runtime-Verifiable Agentic Framework for Reliable Branching Visual Novel Script Generation. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. 2026;22(1):424-433. doi:10.1609/aiide.v22i1.43600

BibTeX
@article{Wei2026,
  author = {Wei, Chuanru and Chen, Kaili},
  title = {GRAIL-VN: A Graph-Grounded Runtime-Verifiable Agentic Framework for Reliable Branching Visual Novel Script Generation},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment},
  year = {2026},
  volume = {22},
  number = {1},
  pages = {424--433},
  doi = {10.1609/aiide.v22i1.43600}
}
Kostenlos starten →