Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{204681,
        author = {Shreeya Daga and Shraddha Gangurde and Samali Rajderkar and Mangesh Gosavi and Vishal Jaiswal},
        title = {Dialogue Generation For Dramas From Transcripts Using Large Language Models},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {7520-7527},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=204681},
        abstract = {High-volume, long-running daily soap television series pose an especially difficult task for large language models (LLMs). Daily soaps require an LLM to generate production-ready dialogue continuously, with the added constraint of maintaining strict episodic continuity and deep consistency in character behaviors throughout multiple episodes. In this survey, we will provide a cohesive framework for using LLMs to coauthor serialized television. We will also examine current advances in LLMs for creative writing for our television framework. The main focus of the survey will be on the technical challenges faced by LLMs when creating scripts for serialized television, including generating large amounts of dialogue (10-15 pages per episode), maintaining coherent narrative structure through hundreds of connected scenes, and faithful representation of each character’s voice during multi-character dialogue scenes. Three essential elements for creating scalable scripted dialog have been identified in our research. First, hierarchical context management for the maintenance of long-term story arc; second, multiagent orchestration for the incorporation of production metadata such as music, sfx, and camera cues; third, human-in-the-loop editing environments that support both non-destructive revision and version control. Our results indicate a paradigmatic shift from one-shot prompt engineering to a modular, context-aware architecture that treats character profiles, narrative history, and technical cues as structured input representations. The major implication of our research is that fine-grained LLM orchestration (i.e., combining long-context modeling, adaptive memory, and collaborative interfaces) is necessary for developing these models from experimental script generators to reliable, highthroughput creative partners capable of supporting the demands of character-driven television storytelling on a perpetual basis.},
        keywords = {Dialogue Generation, Large Language Models, Narrative Continuity, Serialized Television, Script Generation,},
        month = {June},
        }