Maintained by Yingpeng Ma and Yan Ma A curated list of papers on story generation and storytelling in the era of large language models: long-form fiction, screenplays and drama, games, narrative world models, visual stories, and how to evaluate and co-create them. Every paper comes with a one-line summary, and papers we consider essential reading are marked with 🌟. Thank you for the stars! Contributions are very…
This repository collects an extensive list of awesome papers about Story Generation / Storytelling, exclusively focusing on the era of Large Language Models (LLMs).
A curated list of papers on story generation and storytelling in the era of large language models: long-form fiction, screenplays and drama, games, narrative world models, visual stories, and how to evaluate and co-create them. Every paper comes with a one-line summary, and papers we consider essential reading are marked with 🌟.
Thank you for the stars! Contributions are very welcome: open an issue or PR for missing papers or mistakes. Contact: mayingpeng33 [AT] gmail [DOT] com
📰 News
[2026-09] 🎉 Major update: 199 papers, a new taxonomy, one-line summaries and 🌟 must-read picks.
[2026-05] 🔥 Our paper on long-horizon consistency in interactive narratives is accepted to ICML 2026! See it here.
🗺️ Overview
Beyond Text: interactive drama, games, narrative world models, screenplays, and visual stories.
Text Stories: planning, coherence, characters, creativity, and training for written stories.
Evaluation: benchmarks, metrics, and analyses of written stories.
Co-creation: tools for creators, and studies of how people write with AI.
Surveys: overviews of the whole field.
Each paper appears exactly once. Human-centered systems and studies go to Co-creation; work on other media goes to Beyond Text (including its evaluation); remaining work on written stories goes to Evaluation or to the Text Stories topic it mainly addresses. Visual work is included only when it operates at the story level (plot, script, shot planning, narrative reasoning), not when it only improves rendering quality or character consistency. Within a section, papers are sorted by year, with 🌟 must-reads first.
Data table for the chart (2026 counts through September; the 4 surveys are not shown)
How to read an entry: venue · citation count (refreshed weekly) · 🌟 must-read · title · [paper] · GitHub stars of the official code, when available, followed by authors and a one-line summary.
Venue colors:
🎭 Beyond Text
🎪 Interactive Drama
🌟 Can LLM Agents Stick to the Script? A Benchmark for Long-Horizon Consistency in Interactive Narratives[paper] Yingpeng Ma, Jianhao Yan, Bei-Ning Shi, Karim Kam, Runnan Wang, Xue-Bo Liu, Yulong Chen, Yue Zhang, Derek F. Wong
Introduces a 100-environment benchmark on whether LLM narrators keep story commitments under user interventions; even GPT-5.2 survives only 42% after 20 turns.
NARRA-Gym for Evaluating Interactive Narrative Agents[paper] Yue Huang, Yu-Chen Ma, Jiayi Ye, Wen-Jie Wang, Zi-Peng Ling, Xing Hu, Yuexing Hao, Zi-Chen Chen, Zhangchen Xu, Yun-Hong He, et al.
Introduces an executable environment growing emotional seeds into full interactive story episodes, showing fluent LLMs still fail on robustness and personalization.
AdaMARP: An Adaptive Multi-Agent Interaction Framework for General Immersive Role-Playing[paper] Zhenhua Xu, Dongsheng Chen, Shuo Wang, Jian Li, Chengjie Wang, Meng Han, Ya-Biao Wang
Proposes a multi-agent role-play framework whose scene manager selects speakers, switches scenes, and introduces roles, with training data and a benchmark.
HAMLET: A Hierarchical and Adaptive Multi-Agent Framework for Live Embodied Theatrics[paper][dataset] Shu-Fan Jiang, Si-Zhou Chen, Chios Chen, Chi Zhang, Xiao-Lei Zhang, Xue-Long Li
Builds HAMLET, a multi-agent framework that turns a topic into a narrative blueprint and performs live embodied theatre with adaptive actor agents.
🌟 Towards Enhanced Immersion and Agency for LLM-based Interactive Drama[paper] Hongqiu Wu, Weiqi Wu, Tianyang Xu, Jiameng Zhang, Hai Zhao
Proposes Playwriting-guided Generation and Plot-based Reflection to improve player immersion and agency in LLM-based interactive drama.
🌟 CharacterBox: Evaluating the Role-Playing Capabilities of LLMs in Text-Based Virtual Worlds[paper] Lei Wang, Jian-Xun Lian, Yi Huang, Yanqi Dai, Haoxuan Li, Xu Chen, Xing Xie, Ji-Rong Wen
Introduces a simulation sandbox with character and narrator agents that produces behavior trajectories for fine-grained evaluation of LLM role-playing.
OPEN-THEATRE: An Open-Source Toolkit for LLM-based Interactive Drama[paper] Tianyang Xu, Hongqiu Wu, Weiqi Wu, Hai Zhao
Releases Open-Theatre, an open-source toolkit for LLM interactive drama with multi-agent architecture and hierarchical retrieval-based memory for coherent long-term behavior.
RolePlot: A Systematic Framework for Evaluating and Enhancing the Plot-Progression Capabilities of Role-Playing Agents[paper] Pinyi Zhang, Si-Yu An, Lingfeng Qiao, Yi-Fei Yu, Jing-Yang Chen, Jie Wang, Di Yin, Xing Sun, Kai Zhang
Proposes a plot-progression dataset and method for role-playing agents, detecting an LLM embedding trigger subspace to prompt timely plot advances.
🌟 From Role-Play to Drama-Interaction: An LLM Solution[paper] Weiqi Wu, Hongqiu Wu, Lai Jiang, Xing-Chen Liu, Jiale Hong, Haizhen Zhao, Min Zhang
Defines LLM-based interactive drama and trains a drama LLM using Narrative Chain control, Auto-Drama script synthesis, and Sparse Instruction Tuning.
NarrativePlay: An Automated System for Crafting Visual Worlds in Novels for Role-Playing[paper] Run-Cong Zhao, Wenjia Zhang, Jiazheng Li, Lixing Zhu, Yanran Li, Yulan He, Lin Gui
Presents a demo system that lets users role-play a novel character in LLM-generated narrative environments with generated visuals and speech.
🎲 Games
When Stories Evolve: Benchmarking LLM Storytelling Across Agent Architectures in Open-Ended World Simulations[paper] Yuqi Chen, Sixuan Li, Yunfeng Cai, Xueai Li, Kaiwen Yan, Ying Li
Introduces a process benchmark for storytelling in evolving world simulations, finding generation length, canonical consistency, and narrative richness are distinct, competing capacities.
Generating Clue-Driven Investigative Game Narratives with Large Language Models[paper] Vikram Kumaran, A. Smith, Wookhee Min, Randall Spain, Bradford W. Mott, James C. Lester
Builds an LLM framework that generates solvable clue-driven investigative 3D game episodes around a deductive solution model guiding characters, clues, and dialogue.
IVIE: A Neuro-symbolic Approach to Incremental and Validated Generation of Interactive Fiction Worlds[paper] Micaela Vaucher, Santiago Silveira, Santiago Góngora, Luis Chiruzzo
Generates playable interactive fiction worlds in four incremental stages, letting LLMs make creative choices while symbolic validation keeps world state coherent.
Guiding, Not Railroading: Design and Evaluation of a Multi-Agent System for Narrative Redirection in Role-playing Games[paper] Nicolai Hejlesen Jørgensen, Sarmilan Tharmabalan, Ilhan Aslan, Nicolai Brodersen Hansen, Timothy Merritt
Builds a multi-agent RPG game master with a narrative graph and tests six redirection strategies; players prefer in-world redirection over hard denials.
STORY2GAME: Generating (Almost) Everything in an Interactive Fiction Game[paper] E. Zhou, Shreyas Basavatia, M. Siam, Zexin Chen, Mark O. Riedl
Builds STORY2GAME, which generates a story, populates a world, and writes action code from LLM-derived preconditions and effects for playable interactive fiction.
NarrativeGenie: Generating Narrative Beats and Dynamic Storytelling with Large Language Models[paper] Vikram Kumaran, Jonathan Rowe, James C. Lester
Builds NarrativeGenie, which turns a designer’s story overview into a partially ordered event graph of narrative beats that adapts to player actions.
PANGeA: Procedural Artificial Narrative Using Generative AI for Turn-Based, Role-Playing Video Games[paper] Stephanie Buongiorno, Lawrence J. Klinkert, Zixin Zhuang, Tanishq Chawla, Corey Clark
Builds a system with memory, validation, and a Unity plug-in that keeps LLM-generated RPG content consistent with designer rules despite free-form input.
Word2World: Generating Stories and Worlds through Large Language Models[paper] Muhammad Umair Nasir, Steven James, Julian Togelius
Builds Word2World, which prompts LLMs to write a story, extract narrative elements, and place tiles to produce playable game worlds without fine-tuning.
Generating Role-Playing Game Quests With GPT Language Models[paper] Susanna Värtinen, Perttu Hämäläinen, C. Guckelsberger
Fine-tunes GPT-2 on a released dataset of 978 RPG quests, finding about one in five generated quest descriptions acceptable to players.
Ontologically Faithful Generation of Non-Player Character Dialogues[paper] Nathaniel Weir, Ryan Thomas, Randolph D’Amore, Kellie Hill, Benjamin Van Durme, Harsh Jhamtani
Introduces KNUDGE, a dataset from The Outer Worlds requiring lore-faithful, quest-revealing NPC dialogue trees, with supervised and in-context baselines leaving headroom.
🌟 SceneCraft: Automating Interactive Narrative Scene Generation in Digital Games with Large Language Models[paper] Vikram Kumaran, Jonathan Rowe, Bradford W. Mott, James C. Lester
Proposes SceneCraft, an LLM framework that automates NPC interaction scenes to unfold authored plot events in narrative-centered games.
🌟 Language as Reality: A Co-Creative Storytelling Game Experience in 1001 Nights using Generative AI[paper] Yuqian Sun, Zhouyi Li, Ke Fang, Chang Hee Lee, A. Asadipour
Presents 1001 Nights, a game where spoken keywords in co-created LLM tales materialize as in-game items, proposing the notion of AI-native games.
🌟 FIREBALL: A Dataset of Dungeons and Dragons Actual-Play with Structured Game State Information[paper] Andrew Zhu, Karmanya Aggarwal, Alexander H. Feng, Lara J. Martin, Chris Callison-Burch
Releases a dataset of about 25,000 real Discord D&D sessions with true game state, showing state information improves LLM game-turn generation.
Location-Aware Adaptation of Augmented Reality Narratives[paper] Wan-Wan Li, Changyang Li, Minyoung Kim, Haikun Huang, L. Yu
Proposes an optimization approach that assigns real-world locations to AR story events and synthesizes a navigation graph across story branches.
Personalized Quest and Dialogue Generation in Role-Playing Games: A Knowledge Graph- and Language Model-based Approach[paper] Trevor Ashby, Braden K Webb, G. Knapp, John Searle, Nancy Fulda
Proposes a player-centered RPG quest and dialogue generator grounding content in a hand-crafted knowledge base and an LLM, approaching hand-crafted quest quality.
I Cast Detect Thoughts: Learning to Converse and Guide with Intents and Theory-of-Mind in Dungeons and Dragons[paper] Pei Zhou, Andrew Zhu, Jennifer Hu, J. Pujara, Xiang Ren, Chris Callison-Burch, Yejin Choi, Prithviraj Ammanabrolu
Trains a Dungeon Master model with RL that rewards guidance whose intent matches theory-of-mind predictions of player actions in D&D.
🌐 World Models
FilmWorld: Agentic Novel-to-Film Generation through Dynamic Cinematic World Modeling[paper] Jia-Long Zuo, Haotong Zuo, Shiwei Zhang, Xiang Wang, Chen Li, Nong Sang, Chang-Xin Gao, Xiang Bai
Frames novel-to-film generation as building a persistent cinematic world model from prose, then rendering long multi-scene films from it.
EvolvingWorld: An Open-Schema Framework for Co-Evolving Role-Play Agents and World Model in Interactive Literary World[paper] Qing Zong, Yue (Sophie) Guo, Mengxi Yang, Yiwen Guo, Yangqiu Song
Models interactive literary worlds as long-horizon co-evolution of characters and world state, with an open-schema framework and benchmark.
ReactiveGWM: Steering NPC in Reactive Game World Models[paper] Zeqing Wang, Dan Chen, Zhaohu Xing, Zizhao Tong, Yinhan Zhang, Xingyi Yang, Ye-Ying Jin
Decouples player control from NPC behavior in a game world model, so text prompts can steer how NPCs react to the player.
Reformulates multi-shot video generation as causal next-shot prediction, letting users steer an unfolding story in real time via streaming prompts.
🌟 Unbounded: A Generative Infinite Game of Character Life Simulation[paper] Jialu Li, Yuanzhen Li, Neal Wadhwa, Y. Pritch, David E. Jacobs, Michael Rubinstein, Mohit Bansal, Nataniel Ruiz
Builds a generative infinite game in which players raise an autonomous character in an LLM-driven, image-generated world with open-ended, emergent mechanics.
AnimeGamer: Infinite Anime Life Simulation with Next Game State Prediction[paper] Junhao Cheng, Yu-Ying Ge, Yi-Xiao Ge, Jing Liao, Shan Ying
Turns anime film characters into playable agents for open-ended life simulation, predicting multimodal game states to keep the generated world consistent.
🎞️ Screenplays
One Sentence, One Drama: Personalized Short-Form Drama Generation via Multi-Agent Systems[paper] Yu-Fei Shi, Wei-Long Yan, Naixuan Huang, Yucheng Chen, Chenyu Zhang, Tao He, Si Yong Yeo, Ming Li
Builds a hierarchical multi-agent pipeline turning a one-sentence idea into a short drama via debate-based scripting, 3D-grounded first frames, and reviewer loops.
Text-to-Stage: Spatial Layouts from Long-form Narratives[paper] Jefferson Hernandez, Swarnadeep Saha, Chenxi Whitehouse, Sanjeel Parekh, Calvin Murdock, Yuliang Li, W. O. Brimijoin, V. Ithapu, I. Ananthabhotla
Introduces the task of inferring stage layouts and movements from narrative text, with a dramaturgy-based evaluation suite and rejection-SFT plus GRPO training.
COMIC: Agentic Sketch Comedy Generation[paper] Susung Hong, Brian Curless, Ira Kemelmacher-Shlizerman, Steven M. Seitz
Builds an automated agent population mimicking studio roles to produce sketch comedy videos, using LLM critics aligned with YouTube viewer preferences.
DramaBench: A Six-Dimensional Evaluation Framework for Drama Script Continuation[paper] Shijian Ma, Yun-Chien Huang, Yan Lin
Introduces a drama script continuation benchmark scoring six dimensions via rules and LLM labeling, evaluating eight LLMs on 1,103 scripts.
Beyond Direct Generation: A Decomposed Approach to Well-Crafted Screenwriting with LLMs[paper] Hang Lei, Shengyi Zong, Zhaoyan Li, Ziren Zhou, Hao Liu, Liang Yu
Decouples screenplay writing into outline-to-prose then prose-to-screenplay stages with hybrid data synthesis, winning 75% against strong baselines per professional screenwriters.
CML-Bench: A Framework for Evaluating and Enhancing LLM-Powered Movie Scripts Generation[paper] Mingzhe Zheng, Dingjie Song, Guanyu Zhou, Jun You, Jia-Hao Zhan, Xuran Ma, Xin-Yuan Song, Ser-Nam Lim, Qi-Feng Chen, Harry Yang
Introduces a movie script benchmark scoring dialogue coherence, character consistency, and plot reasonableness, plus an instruction-based prompting strategy for better scripts.
🌟 IBSEN: Director-Actor Agent Collaboration for Controllable and Interactive Drama Script Generation[paper] Senyu Han, Lu Chen, Li-Min Lin, Zhen Xu, Kai Yu
Proposes IBSEN, where a director agent steers actor agents and human players toward plot objectives to generate controllable drama scripts.
🌟 HoLLMwood: Unleashing the Creativity of Large Language Models in Screenwriting via Role Playing[paper] Jing Chen, Xinyu Zhu, Cheng Yang, Chufan Shi, Ya-Dong Xi, Yuxiang Zhang, Junjie Wang, Jiashu Pu, Rongsheng Zhang, Yu-Jiu Yang, et al.
Builds HoLLMwood, a screenwriting framework assigning LLMs Writer, Editor, and role-playing Actor roles to enrich characters and plots in generated screenplays.
🌟 Co-Writing Screenplays and Theatre Scripts with Language Models: An Evaluation by Industry Professionals[paper] Piotr Wojciech Mirowski, K. Mathewson, Jaylen Pittman, Richard Evans
Builds Dramatron, which hierarchically prompts LLMs to co-write scripts and screenplays, evaluated in a study with 15 theatre and film professionals.
🖼️ Visual2Story
Generating Visual Stories with Grounded and Coreferent Characters[paper] Danyang Liu, Mirella Lapata, Frank Keller
Presents a character-centric visual storytelling model trained on VIST enriched with visual and textual coreference chains, plus metrics for character richness.
🌟 StoryLLaVA: Enhancing Visual Storytelling with Multi-Modal Large Language Models[paper] Li Yang, Zhihao Xiao, Wen-Xin Huang, Xian Zhong
Proposes a visual storytelling MLLM trained with a topic-driven narrative optimizer for data refinement and preference-based ranked story sampling for alignment.
VIST-GPT: Ushering in the Era of Visual Storytelling with LLMs?[paper] M. Gado, Towhid Taliee, M. Memon, Dmitry Ignatov, R. Timofte
Adapts large multimodal models to visual storytelling on VIST and advocates reference-free metrics RoViST and GROOVIST over BLEU-style evaluation.
From Panels to Prose: Generating Literary Narratives from Comics[paper] Ragav Sachdeva, Andrew Zisserman
Builds a system that converts manga into literary prose for visually impaired readers, introducing the Magiv3 comic-understanding model and annotated panel captions.
Not (yet) the whole story: Evaluating Visual Storytelling Requires More than Measuring Coherence, Grounding, and Repetition[paper] Aditya K Surikuchi, Raquel Fernández, Sandro Pezzelle
Proposes a human-likeness metric over visual grounding, coherence, and repetition, finding a small upgraded TAPM rivals LLaVA, yet good stories need more.
Synchronized Video Storytelling: Generating Video Narrations with Structured Storyline[paper] Dingyi Yang, Chunru Zhan, Ziheng Wang, Biao Wang, Tiezheng Ge, Bo Zheng, Qin Jin
Introduces synchronized video storytelling, generating clip-aligned narrations of fitting length, with the E-SyncVidStory dataset and a storyline-guided VideoNarrator framework.
SCO-VIST: Social Interaction Commonsense Knowledge-based Visual Storytelling[paper] E. Wang, Caren Han, Josiah Poon
Proposes a visual storytelling framework that builds a social-commonsense plot graph from images and derives storylines via weighted shortest paths with Floyd-Warshall.
🌟 Visual Writing Prompts: Character-Grounded Story Generation with Curated Image Sequences[paper] Xudong Hong, A. Sayeed, K. Mehra, Vera Demberg, B. Schiele
Introduces a dataset of about 2K curated movie-shot sequences with 12K character-grounded crowdsourced stories, plus a coherence-driven character-based generation baseline.
DiffuVST: Narrating Fictional Scenes with Global-History-Guided Denoising Models[paper] Shengguang Wu, Mei Yuan, Qi Su
Proposes a non-autoregressive diffusion model that generates visual story narrations for fictional image sequences with bidirectional history guidance, improving speed and diversity.
Sound of Story: Multi-modal Storytelling with Audio[paper] Jaeyeon Bae, Seokhoon Jeong, Seokun Kang, Namgi Han, Jae-Yon Lee, Hyounghun Kim, Taehwan Kim
Introduces Sound of Story, a dataset of 27K stories pairing image-text sequences with background audio, plus cross-modal retrieval and audio generation benchmarks.
GROOViST: A Metric for Grounding Objects in Visual Storytelling[paper] Aditya K Surikuchi, Sandro Pezzelle, Raquel Fernández
Proposes a modular, interpretable metric for visual storytelling that measures how well stories are grounded in image entities, handling temporal misalignment.
Visual Storytelling with Question-Answer Plans[paper] Danyang Liu, Mirella Lapata, Frank Keller
Proposes visual storytelling that feeds images as a visual prefix to a pretrained language model and plans with question-answer blueprints.
Attractive Storyteller: Stylized Visual Storytelling with Unpaired Text[paper] Dingyi Yang, Qin Jin
Introduces stylized visual storytelling and a memory-augmented multitask model trained with unpaired style text to generate styled stories from photo streams.
Multimodal Event Transformer for Image-guided Story Ending Generation[paper] Yucheng Zhou, Guodong Long
Proposes an event-graph reasoning transformer for image-guided story ending generation, with cross-modal fusion, a multimodal injector, and incoherence detection.
🌄 Story2Visual
🌟 MAViS: A Multi-Agent Framework for Long-Sequence Video Storytelling[paper] Qian Wang, Zi-Qi Huang, Ruoxi Jia, Paul E. Debevec, Ning Yu
Proposes a multi-agent pipeline spanning scripting, shot design, character modeling, keyframes, animation, and audio for long-sequence video storytelling.
Better Call CineCrew: Consistent Ultra-Long Narrative-to-Film Generation[paper] Jiaben Chen, Si Dong, Qinhong Zhou, Raine Ma, Zhi-Yang Dou, Wojciech Matusik, Chuang Gan
Proposes a multi-agent orchestration layer built on FilmDSL, a film-specific language making shot, continuity, and persona constraints explicit for long script-to-video generation.
Learning Long-form Movie Prior via Large Language Models.[paper] Jin-Heng Xie, Jia-Jun Feng, M. Shou
Represents movies as text and bounding-box or keypoint tokens and curates Storyboard20K, letting LLMs learn movie priors to sample storyboards.
Builds a deployed storyboarding system that learns directing rules from expert examples, evolves them via attribution feedback, and releases the PROSE dataset.
AniMaster: From Story Texts to Animated Videos via Cinematic Script Generation and Interactive Authoring[paper] Ruiqi Yu, De-Kun Qian, Jia-Le Xu, Si-Zhe Cheng, Yize Li, Xiang-yang Wu, Zhiguang Zhou, Wei Chen, Yong Wang
Builds an authoring tool that expands brief story texts into cinematic scripts, then animated videos, guided by a three-layer design framework.
MangaFlow: An End-to-End Agentic Framework for Controllable Story to Manga Generation[paper] Mu-Yao Wang, Ze-Ke Xie, Yanhao Chen, Lixin Xiu, Hideki Nakayama
Proposes an agentic story-to-manga framework decomposing creation into planning, grounding, layout, rendering, composition, and lettering for controllable page generation.
BOOKAGENT: Orchestrating Safety-Aware Visual Narratives via Multi-Agent Cognitive Calibration[paper] Bo Gao, Chang Liu, Yu-Yang Miao, Siyuan Ma, S. Lim
Proposes a safety-aware multi-agent framework for end-to-end illustrated storybook generation with page-level text-image calibration and global consistency repair.
CANVAS: Continuity-Aware Narratives via Visual Agentic Storyboarding[paper] I. Mondal, Yi-Wen Song, Mihir Parmar, Palash Goyal, J. Boyd-Graber, Tomas Pfister, Yale Song
Proposes a multi-agent storyboarding framework that plans character, background, and location continuity, plus a new long-range consistency benchmark.
LogiStory: A Logic-Aware Framework for Multi-Image Story Visualization[paper] Chutian Meng, Fan Ma, Chi Zhang, Jiaxu Miao, Yi Yang, Yue-Ting Zhuang
Proposes a multi-agent story visualization framework that grounds roles, extracts causal chains, and verifies consistency to model visual logic explicitly.
Introduces emotion-aware visual story generation and a two-stage framework combining agent-based planning with region-aware generation for emotional, subject-consistent image sequences.
Narrative Weaver: Towards Controllable Long-Range Visual Consistency with Multi-Modal Conditioning[paper] Zhengjian Yao, Yong-Zhi Li, Xinyu Gao, Quan Chen, Peng Jiang, Yan Lu
Combines an MLLM narrative planner with a memory-bank control module for long consistent visual sequences and releases a 330K-image e-commerce storyboard dataset.
MUSE: A Multi-agent Framework for Unconstrained Story Envisioning via Closed-Loop Cognitive Orchestration[paper] Wenzhang Sun, Zhenyu Wang, Zhang-Chi Hu, Chun-Feng Wang, Hao Li, Wei Chen
Proposes a multi-agent plan-execute-verify-revise loop for long audio-visual stories from short prompts, plus a reference-free evaluation protocol.
🌟 SEED-Story: Multimodal Long Story Generation with Large Language Model[paper] Shuai Yang, Yu-Ying Ge, Yang Li, Yu-Kang Chen, Yixiao Ge, Shan Ying, Ying-Cong Chen
Proposes SEED-Story, an MLLM generating interleaved text and consistent images for long stories via multimodal attention sinks, and releases the StoryStream dataset.
Generating Storytelling Images with Rich Chains-of-Reasoning[paper] Xiujie Song, Qi Jia, Shota Watanabe, Xiao-Yi Pang, Ruijie Chen, Mengyue Wu, Ke Zhu
Defines storytelling image generation with chains of visual reasoning clues and proposes an LLM-plus-text-to-image pipeline with dedicated evaluation metrics.
From Outline to Detail: An Hierarchical End-to-end Framework for Coherent and Consistent Visual Novel Generation and Assembly[paper] Yilin Zhang, Yanyan Wei, Zhao Zhang, Jicong Fan, Haijun Zhang, Shui-Cheng Yan
Proposes an outline-guided pipeline that generates and assembles executable visual novels, using vision-LLM self-correction for cross-modal consistency and script validation.
LLMs Behind the Scenes: Enabling Narrative Scene Illustration[paper] Melissa Roemmele, John Joon Young Chung, Taewook Kim, Yuqian Sun, Alex Calderwood, Max Kreminski
Uses LLMs to prompt text-to-image models for narrative scene illustration and releases SceneIllustrations, a dataset of pairwise human quality judgments.
AniMaker: Multi-Agent Animated Storytelling with MCTS-Driven Clip Generation[paper] Haoyuan Shi, Yunxin Li, Xinyu Chen, Long-Yue Wang, Bao-Tian Hu, Min Zhang
Proposes AniMaker, a multi-agent animation framework using MCTS-driven multi-candidate clip generation and the AniEval evaluator to produce story-coherent long videos from text.
VinaBench: Benchmark for Faithful and Consistent Visual Narratives[paper] Silin Gao, Sheryl Mathew, Li Mi, Sepideh Mamooler, Mengjie Zhao, Hiromi Wakaki, Yuki Mitsufuji, Syrielle Montariol, Antoine Bosselut
Introduces a benchmark annotating commonsense and discourse constraints in visual narratives, with metrics for consistency and text alignment of generated image sequences.
MM-StoryAgent: Immersive Narrated Storybook Video Generation with a Multi-Agent Paradigm across Text, Image and Audio[paper] Xuenan Xu, Jiahao Mei, Chenliang Li, Yuning Wu, Ming Yan, Shaopeng Lai, Ji Zhang, Mengyue Wu
Proposes a multi-agent framework combining LLMs with image, speech, music, and sound tools to generate narrated storybook videos for children.
VISIAR: Empower MLLM for Visual Story Ideation[paper] Zhaoyang Xia, Somdeb Sarkhel, Md Mehrab Tanjim, Stefano Petrangeli, Ishita Dasgupta, Yuxiao Chen, Jinxuan Xu, Di Liu, Saayan Mitra, Dimitris N. Metaxas
Introduces visual story ideation, arranging visual assets into storylines, with an MLLM framework using a story graph and a VTravel benchmark.
Multimodal Persona Based Generation of Comic Dialogs[paper] Harsh Agrawal, A. Mishra, Manish Gupta, M. -
Introduces multimodal persona-based comic dialogue generation with a 54K-strip dataset and an architecture that generates next-panel dialogues.
✍️ Text Stories
🗺️ Planning
LLM-Driven MCTS for Conditional Story Generation via Logic-Guided Evidence Tree Optimization[paper] Hongyan Wu, Zhiliang Tian, Zhen Huang, Nankai Lin, Yi-Ping Song, Zhihua Wen, Menglong Lu, Feng Liu, Dongsheng Li
Proposes a plug-and-play MCTS planner that builds logic-validated evidence chains for retrieval-based conditional story generation to reduce incoherence and thematic drift.
Proposes PLOTTER, which runs an Evaluate-Plan-Revise cycle on event and character graphs to fix causality and structure before generating full narrative text.
BiT-MCTS: A Theme-based Bidirectional MCTS Approach to Chinese Fiction Generation[paper] Zhaoyi Li, Xu Zhang, Xiaojun Wan
Generates Chinese fiction by writing the climax first, then expanding plot backward and forward with bidirectional MCTS inspired by Freytag’s Pyramid.
Lightweight Latent Reasoning for Narrative Tasks[paper] Alexander Gurung, Esmeralda S. Whitammer, Mirella Lapata
Proposes a lightweight reasoning projector producing continuous latent tokens that RL policies toggle, cutting reasoning length on plot-hole detection and chapter generation.
🌟 Generating Long-form Story Using Dynamic Hierarchical Outlining with Memory-Enhancement[paper] Qianyue Wang, Jinwu Hu, Zhengpin Li, Yufeng Wang, Daiyuan Li, Yu Hu, Mingkui Tan
Proposes DOME, which fuses planning and writing through dynamic hierarchical outlines and uses a memory module to reduce contradictions in long stories.
🌟 Agents’ Room: Narrative Generation through Multi-step Collaboration[paper] Fantine Huot, Reinald Kim Amplayo, J. Palomaki, Alice Shoshana Jakobovits, Elizabeth Clark, Mirella Lapata
Proposes Agents’ Room, which splits fiction writing into subtasks for specialized agents, and releases the Tell Me A Story dataset and evaluation.
StoryWriter: A Multi-Agent Framework for Long Story Generation[paper] Haotian Xia, Hao Peng, Yunjia Qi, Bin Xu, Juan-Zi Li, Hou Lei, Xiaozhi Wang
Proposes a multi-agent long story framework and uses it to build a 6,000-story dataset for fine-tuning Llama3.1-8B and GLM4-9B.
STORYTELLER: An Enhanced Plot-Planning Framework for Coherent and Cohesive Story Generation[paper] Jiaming Li, Yu-Kun Chen, Ziqiang Liu, Minghuan Tan, Lei Zhang, Yunshui Li, Run Luo, Long-Ze Chen, Jing Luo, A. Argha, et al.
Proposes a plot-planning approach using SVO-triplet plot nodes plus interacting storyline and narrative entity knowledge graph modules for coherent story generation.
Ex3: Automatic Novel Writing by Extracting, Excelsior and Expanding[paper] H. Lei, Jiaming Guo, Guanhua He, Xi-Shan Zhang, Rui Zhang, Shaohui Peng, Shaoli Liu, Tianshi Chen
Proposes Ex3, which extracts structure from raw novels to build instruction data, fine-tunes an LLM, and expands tree-like into arbitrarily long novels.
SWAG: Storytelling With Action Guidance[paper] Zeeshan Patel, Karim El-Refai, Jonathan Pei, Tianle Li
Proposes SWAG, framing story writing as search where an auxiliary LLM picks the next action steering the generator toward engaging stories.
Little Red Riding Hood Goes around the Globe: Crosslingual Story Planning and Generation with Large Language Models[paper] E. Razumovskaia, Joshua Maynez, Annie Louis, Mirella Lapata, Shashi Narayan
Introduces crosslingual story generation with planning and a dataset, finding three-act plans yield more coherent, interesting, controllable stories across languages.
🌟 DOC: Improving Long Story Coherence With Detailed Outline Control[paper] Kevin Yang, D. Klein, Nanyun Peng, Yuan-Dong Tian
Improves long-story plot coherence by generating a detailed hierarchical outline and a controller that keeps drafted passages aligned with outline details.
🧵 Coherence
FossilWriter: Learning Hypergraph World Models with Latent Narratives for Creative Story Generation[paper] Heng Zhang, Yi-Hao Zhong, Lubin Gan, Zhihe Chen, Tianyi Zhang, Jing Liu, Jin Huang
Grows a hypergraph world model whose unresolved elements seed latent narratives, improving plot coherence and reducing long-range factual conflicts in story generation.
Narrative World Model: Narratology-Grounded Writer Memory for Long-Form Fiction[paper] M. Saifullah, Thomas Kornmaier, Taaha Kazi, Vasu Sharma, A. Kanade, Aanand Kumar Yadav
Proposes a writer-memory system pairing a narratology-typed temporal state graph with hybrid retrieval, outperforming Graphiti/Zep and GraphRAG on multi-hop story questions.
ConWriter: Transition-Constrained Stateful Long-Form Story Generation with Lightweight Neuro-Symbolic Consistency Control[paper] Jindong Li, Yang Yang, Zihao Liu, Yutao Yue, Meng-Lin Yang
Proposes a training-free scene-by-scene writer that tracks symbolic story states, checks narrative transitions, and uses uncertainty signals to repair inconsistencies.
Octopus: Entropy-Controlled Science Fiction Literature Generation with Persistent Memory-Context Binding[paper] Xu Wang, Jiaju Kang, Puyu Han, Zeyu Ai, Luqi Gong
Proposes Octopus, combining entropy regulation via narrative divergence thresholds with hierarchical memory of characters, plots, and scientific rules for long sci-fi generation.
SCORE: Story Coherence and Retrieval Enhancement for AI Narratives[paper] Qiang Yi, Yang-Fan He, Jian-Hui Wang, Xin-Yuan Song, Shi-Yao Qian, Xin-Hang Yuan, Yi Xin, Yi-Jin Wang, Jingqun Tang, Yuchen Li, et al.
Proposes SCORE, which tracks key item states and episode summaries and uses retrieval-augmented generation to detect and fix inconsistencies in LLM-generated stories.
MLD-EA: Check and Complete Narrative Coherence by Introducing Emotions and Actions[paper] Jin-Ming Zhang, Yun-Fei Long
Proposes MLD-EA, which uses LLMs with emotion and action cues to detect missing logic in narratives and generate sentences that restore coherence.
FACTTRACK: Time-Aware World State Tracking in Story Outlines[paper] Zhiheng Lyu, Kevin Yang, Lingpeng Kong, Daniel Klein
Proposes FACTTRACK, which decomposes events into atomic facts with time-aware validity intervals to track world state and detect contradictions in story outlines.
🌟 RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text[paper] Wangchunshu Zhou, Y. Jiang, Peng Cui, Tiannan Wang, Zhenxin Xiao, Yifan Hou, Ryan Cotterell, Mrinmaya Sachan
Proposes RecurrentGPT, which simulates LSTM-style recurrence with natural-language long- and short-term memories so LLMs can interactively generate arbitrarily long text.
🧑🤝🧑 Characters
ANIMASK: What the Model Contributes to Role Play in Simulated Story Worlds[paper] Xiu-Cheng Zhang, Zhuo-Ning Xu, Han-Jun Luo, Yankai Chen, Hanan Salam, Xue Liu
Replays story worlds from freeze points with and without personas, finding actor LLMs push characters toward cautious, flatter outcomes than canon.
Staying In Character: Perspective-Bounded Memory For Book-Based Role-Playing Agents[paper] Xushuo Tang, Junhe Zhang, Zi-Han Yang, Yi-Fu Tang, Sichao Li, Longbin Lai, Zheng-Yi Yang
Proposes a three-layer, perspective-bounded memory for book-based role-playing agents that prevents characters using unknown facts, with a 4,386-question knowledge-boundary benchmark.
EvoSpark: Endogenous Interactive Agent Societies for Unified Long-Horizon Narrative Evolution[paper] Shiyu He, Min-Chi Kuang, Mengxian Wang, Bin Hu, Tingxiang Gu
Proposes a multi-agent framework with stratified narrative memory and role-location-plot alignment to sustain coherent, open-ended long-horizon story evolution.
Deriving Character Logic from Storyline as Codified Decision Trees[paper] Letian Peng, Kun Zhou, Longfei Yun, Yu-Peng Hou, Jingbo Shang
Induces executable, interpretable decision trees of validated scene-conditioned behavior rules from narrative data to ground role-playing agents more reliably.
StoryBox: Collaborative Multi-Agent Simulation for Hybrid Bottom-Up Long-Form Story Generation Using Large Language Models[paper] Zehao Chen, Rong Pan, Hao-Ran Li
Proposes bottom-up long-form story generation in which multi-agent sandbox simulation yields emergent events that form coherent stories exceeding 10,000 words.
🌟 BookWorld: From Novels to Interactive Agent Societies for Creative Story Generation[paper] Yi-Ting Ran, Xintao Wang, Tian Qiu, Jiaqing Liang, Yanghua Xiao, Deqing Yang
Builds BookWorld, which simulates multi-agent societies from established novels’ characters and worldviews to generate creative stories faithful to the source books.
Steering Narrative Agents Through a Dynamic Cognitive Framework for Guided Emergent Storytelling[paper] Chen Yang, M. Gross, R. Wampfler
Proposes a cognitive agent framework where tensions between agents’ beliefs and ideal worlds drive actions, steering emergent stories toward authored storylines.
🌟 StoryVerse: Towards Co-authoring Dynamic Plot with LLM-based Character Simulation via Narrative Planning[paper] Yi Wang, Qian Zhou, David Ledo
Proposes StoryVerse, where authors write abstract acts that LLM narrative planning turns into character actions, balancing authorial intent with emergent game plots.
🎨 Creativity
MUSE: A Theory-Harnessed Story Engine for Vibe Narrativizing[paper] Jian-Xiang Ma, Xiaocui Yang, Da-Ling Wang, Yue-Song Hou, Ming-Fu Zhang, Yi-Chen Gao, Jun-Zhao Huang
Builds a story engine encoding McKee’s story theory as atomized rules inside an agent harness, improving WritingBench and consistency across four models.
CraftAlign: Feature-Grounded Evaluation and Revision Guidance for AI Stories[paper] Yang Yang, Boyun Xu, Shaofeng Liang, Yun Han, Zining Zhong, Songning Lai, Kaishen Yuan, Yutao Yue
Predicts 304 writing features to score stories against human and AI patterns, turning feature shifts into revision guidance that reduces AI flavor.
StorySpark: Module-wise Evolutionary Search for Story Premise Generation[paper] Yang Yang, Zining Zhong, Qian Cao, Jindong Li, Boyun Xu, Kaishen Yuan, Meng-Lin Yang, Yutao Yue
Proposes module-wise evolutionary search over premise components like persona, event, and twist, producing more original premises that yield better downstream stories.
PlotTwist: A Creative Plot Generation Framework with Small Language Models[paper] A. Thorat, Ravi Kolla, Jyotin Goel, Madhav Kataria, N. Pedanekar
Proposes a framework where sub-3B models generate premise-conditioned plots using an aspect reward model, DPO-aligned MoE generator, and cross-family jury evaluation.
LLM Review: Enhancing Creative Writing via Blind Peer Review Feedback[paper]
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
评论