awesome-llm-story-generation 仓库索引
- 仓库: https://github.com/Picrew/awesome-llm-story-generation
- 维护者: Picrew（即 [[01_lost_in_stories]] / ConStory-Bench 的作者团队）
- 规模: 288 篇，10 大类，覆盖 2022–2026
- 快照时间: 仓库更新于 2026-06-29；本索引整理于 2026-07
说明：抓取时多数条目的具体 arXiv ID 显示为占位符（"arXiv link"），故本索引以标题 + venue/年份为准，需要时按标题检索原文。带 ⭐ 的为本目录已详读的论文。
✅ 已从本合集中挑出 26 篇做了详细解读（长篇生成 / 多智能体 / 叙事技巧 / 评估 / 中文场景五方向），完整分组表见 → README.md，笔记见
papers/。
分类规模
| 分类 | 数量 | 与"文字小说/长文写作"相关度 |
|---|---|---|
| Planning / Decomposition（规划/分解） | 21 | ★★★ 高 |
| Agent Collaboration（多智能体协作） | 6 | ★★★ 高 |
| Sandbox / World Simulation（世界模拟/互动叙事） | 17 | ★★ 中（偏游戏/互动剧） |
| Multimodal（图像/视频/漫画/音频） | 64 | ★ 低（偏视觉生成） |
| Memory & Long-Context Coherence（记忆/长上下文） | 20 | ★★★ 高 |
| Consistency / Controllability（一致性/可控性） | 27 | ★★★ 高 |
| Refinement / Self-Critique（精修/自我批评） | 16 | ★★★ 高 |
| Evaluation / Benchmarks（评估/基准） | 75 | ★★ 中高 |
| Datasets / Surveys / Resources（数据/综述） | 32 | ★★ 中高 |
| Open-source Projects（开源项目） | 10 | ★★ 中 |
A. Planning / Decomposition for Story Generation（21）
- GraphStory: Collaborative Story Writing through Event-Based Narrative Editing (arXiv 2026-06)
- Fabula: Building a Narrative Storytelling Sidekick with the Writers' Community (arXiv 2026-06)
- Towards Human-Level Book-Writing Capability (arXiv 2026-05) ← 书籍级写作，重点
- Planning Beyond Text: Graph-based Reasoning for Complex Narrative Generation (arXiv 2026-04)
- Narrix: Remixing Narrative Strategies from Examples for Story Writing (CHI 2026-04)
- BiT-MCTS: A Theme-based Bidirectional MCTS Approach to Chinese Fiction Generation (arXiv 2026-03) ← 中文小说
- DPWriter: Reinforcement Learning with Diverse Planning Branching for Creative Writing (arXiv 2026-01)
- Codified Foreshadowing-Payoff Text Generation (arXiv 2026-01) ⭐ [[04_codified_foreshadowing_payoff]]
- SceneDecorator: Scene-Oriented Story Generation with Scene Planning and Consistency (arXiv 2025-10)
- Long Story Generation via Knowledge Graph and Literary Theory (arXiv 2025-08)
- STORYTELLER: An Enhanced Plot-Planning Framework for Coherent Story Generation (arXiv 2025-06)
- Can LLMs Generate Good Stories? Insights from a Narrative Planning Perspective (arXiv 2025-06)
- Learning to Reason for Long-Form Story Generation (arXiv 2025-03) — code: Alex-Gurung/ReasoningNCP
- DOME: Dynamic Hierarchical Outlining with Memory-Enhancement (NAACL 2024-12) ⭐ [[03_dome]]
- Ex3: Automatic Novel Writing by Extracting, Excelsior and Expanding (ACL 2024-08)
- Navigating the Path of Writing: Outline-guided Text Generation with LLMs (NAACL 2024-04)
- Creating Suspenseful Stories: Iterative Planning with LLMs (EACL 2024-02)
- Improving Pacing in Long-Form Story Planning (EMNLP Findings 2023-11)
- End-to-End Story Plot Generator (arXiv 2023-10)
- The Next Chapter: A Study of LLMs in Storytelling (arXiv 2023-01)
- DOC: Improving Long Story Coherence With Detailed Outline Control (arXiv 2022-12)
B. Agent Collaboration for Story Writing（6）
- Improving Collaborative Storytelling with a Multi-Agent Framework Based on LLMs (arXiv 2026-05)
- Collaborative Multi-Agent Scripts Generation for Murder Mystery Games (ACL Findings 2026-04)
- A Cognitive Writing Perspective for Constrained Long-Form Text Generation (CogWriter) (arXiv 2025-02) — code: KaiyangWan/CogWriter
- Agents' Room: Narrative Generation through Multi-step Collaboration (ICLR 2024-10)
- HoLLMwood: Unleashing Creativity in Screenwriting via Role Playing (EMNLP Findings 2024-06)
- AutoAgents: A Framework for Automatic Agent Generation (IJCAI 2023-09)
C. Sandbox / World Simulation（17，偏互动叙事/游戏）
- Orchestrated Reality: LLM-Driven World Simulation as Parameterized-Action POMDP (arXiv 2026-06)
- IVIE: Neuro-symbolic Incremental Generation of Interactive Fiction Worlds (ICCC 2026-06)
- BotDirector: Robot Storytelling Across Symmetrical Reality (arXiv 2026-06)
- World-State Transformations for Neuro-symbolic Interactive Storytelling (arXiv 2026-05)
- Material for Thought: Generative AI as an Active Creative Medium (arXiv 2026-05)
- A GenAI Driven Interactive Narrative Serious Game for Stress Relief (arXiv 2026-05)
- A Reflective Storytelling Agent for Older Adults (arXiv 2026-05)
- EvoSpark: Endogenous Interactive Agent Societies for Narrative Evolution (ACL 2026-04)
- StoryBox: Collaborative Multi-Agent Simulation for Bottom-Up Long-Form Story Generation (arXiv 2025-10) — storyboxproject.github.io
- OPEN-THEATRE: Open-Source Toolkit for LLM-based Interactive Drama (arXiv 2025-09)
- HAMLET: Hyperadaptive Agent-based Modeling for Live Embodied Theatrics (arXiv 2025-07)
- STORY2GAME: Generating (Almost) Everything in an Interactive Fiction Game (arXiv 2025-05)
- BookWorld: From Novels to Interactive Agent Societies (arXiv 2025-04) — bookworld2025.github.io
- Towards Enhanced Immersion and Agency for LLM-based Interactive Drama (arXiv 2025-02)
- IBSEN: Director-Actor Agent Collaboration for Drama Script Generation (ACL 2024-07) — code: OpenDFM/ibsen
- StoryVerse: Co-authoring Dynamic Plot with LLM-based Character Simulation (FDG 2024-05)
- Generative Agents: Interactive Simulacra of Human Behavior (arXiv 2023-04) — code: joonspk-research/generative_agents
D. Multimodal（64，偏图像/视频/漫画/音频，此处仅列文字相关的少数）
- R^2: LLM-based Novel-to-Screenplay Generation with Causal Plot Graphs (ICLR 2025-03)
- LongWriter-V: Ultra-Long High-Fidelity Generation in VLMs (arXiv 2025-02) — code: THU-KEG/LongWriter-V
- SEED-Story: Multimodal Long Story Generation (arXiv 2024-07) — code: TencentARC/SEED-Story
- Directing the Narrative: Finetuning for Controlling Coherence and Style (arXiv 2026-03)
- EmoStory: Emotion-Aware Story Generation (arXiv 2026-03)
- （其余 59 篇为视频/图像一致性、多镜头长视频、漫画、音频等，与文字写作弱相关，略）
E. Memory & Long-Context Coherence（20）
- Staying In Character: Perspective-Bounded Memory For Book-Based Role-Playing Agents (arXiv 2026-06)
- Not All Claims Are Equally Risky: FACTOR for Adaptive Verification in Factual Long-Form Generation (arXiv 2026-06) — code: TreeLLi/... 见下
- Storyline Trees: Hierarchical Representations for Long-Form Narratives (arXiv 2026-06) ← 结构表示，重点
- IS-CoT: Breaking the Long-form Generation Collapse via Interleaved Structural Thinking (arXiv 2026-06)
- Narrative Knowledge Weaver: Narrative-Centric RAG for Long-Form Text Understanding (arXiv 2026-06)
- POLARIS: Guiding Small Models to Write Long Stories (arXiv 2026-06) ← 小模型写长文，重点
- Building Reliable Long-Form Generation via Hallucination Rejection Sampling (arXiv 2026-06) — code: TreeLLi/hallucination-rejection-sampling
- Tournament-GRPO: Group-Wise Tournament Rewards for RL in Open-Ended Long-Form Generation (arXiv 2026-05)
- On Stable Long-Form Generation: Benchmarking and Mitigating Length Volatility (arXiv 2026-05)
- Think Before you Write: QA-Guided Reasoning for Character Descriptions in Books (arXiv 2026-04)
- Skeleton-based Coherence Modeling in Narratives (arXiv 2026-04)
- Shifting Long-Context LLMs Research from Input to Output (arXiv 2025-03)
- Language Models can Self-Lengthen to Generate Long Texts (arXiv 2024-10) — code: QwenLM/Self-Lengthen
- LongGenBench: Benchmarking Long-Form Generation in Long-Context LLMs (arXiv 2024-09)
- LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs (arXiv 2024-08) — code: THUDM/LongWriter
- LongLaMP: A Benchmark for Personalized Long-form Text Generation (arXiv 2024-07)
- CHIRON: Rich Character Representations in Long-Form Narratives (EMNLP Findings 2024-06)
- With Greater Text Comes Greater Necessity: Inference-Time Training Helps Long Text Generation (COLM 2024-01)
- LongAlign: A Recipe for Long Context Alignment (arXiv 2024-01) — code: THUDM/LongAlign
- RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text (arXiv 2023-05) — code: aiwaves-cn/RecurrentGPT
F. Consistency / Controllability / Constraint Following（27）
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning (arXiv 2026-06)
- DeSRPA: Decoupled Speech Role-Playing Agent (INTERSPEECH 2026-06)
- Steering Emotional Dynamics for Art Therapy: Controllable Narrative Script Generation (arXiv 2026-06)
- Creative Collision: Directorial Persona Steering and Competition in LLMs (ICML Workshop 2026-06)
- Constrained Semantic Decompression via Persian Proverb-Conditioned Story Generation (arXiv 2026-06)
- Children's English Reading Story Generation via SFT with Controllable Difficulty/Safety (arXiv 2026-05)
- UniCreative: Unifying Long-form Logic and Short-form Sparkle via Reference-Free RL (arXiv 2026-04)
- Noise Steering for Controlled Text Generation (Arabic Educational Story) (arXiv 2026-04)
- Preconditioned Test-Time Adaptation for OOD Debiasing in Narrative Generation (arXiv 2026-03)
- TaleFrame: Interactive Story Generation System with Fine-Grained Control (arXiv 2025-12)
- SCORE: Story Coherence and Retrieval Enhancement for AI Narratives (arXiv 2025-03)
- Whose story is it? Personalizing story generation by inferring author styles (arXiv 2025-02)
- Pastiche Novel Generation: Fan Fiction in Your Favorite Author's Style (arXiv 2025-02)
- CS4: Measuring Creativity by Controlling Story-Writing Constraints (arXiv 2024-10) — code: anirudhlakkaraju/cs4_benchmark
- Crafting Narrative Closures: Zero-Shot with SSM Mamba for Story Ending (arXiv 2024-10)
- MirrorStories: Reflecting Diversity through Personalized Narrative Generation (EMNLP 2024-09)
- FACTTRACK: Time-Aware World State Tracking in Story Outlines (NAACL 2024-07)
- Suri: Multi-constraint Instruction Following for Long-form Text Generation (arXiv 2024-06) — code: chtmp223/suri
- MoPS: Modular Story Premise Synthesis (ACL 2024-06) — code: GAIR-NLP/MoPS
- Measuring Psychological Depth in Language Models (EMNLP 2024-06)
- Guiding and Diversifying LLM Story Generation via Answer Set Programming (ACL Workshop 2024-06)
- Multigenre AI-powered Story Composition (arXiv 2024-05)
- Returning to the Start: Generating Narratives with Related Endpoints (NAACL 2024-04) — code: adbrei/RENarGen
- NarrativeGenie: Generating Narrative Beats and Dynamic Storytelling (AIIDE 2024-01)
- CAT-LLM: Prompting LLMs with Text Style Definition for Chinese Article-style Transfer (arXiv 2024-01)
- Learning to Generate Text in Arbitrary Writing Styles (arXiv 2023-12)
- RLCD: Reinforcement Learning from Contrast Distillation (ICLR 2023-07)
G. Refinement / Self-Critique / Iterative Editing（16）
- OPERA: Aligning Open-Ended Reasoning via Objective Perplexity-based RL (arXiv 2026-06) — code: pangpang-xuan/OPERA
- StoryLens: Preference-Aligned Story Rewriting via Context-Aware Narrative Enrichment (arXiv 2026-05)
- DTO: Differentiable Training Objective for Counterfactual Story Rewriting (arXiv 2026-05)
- R2-Write: Reflection and Revision for Open-Ended Writing with Deep Reasoning (arXiv 2026-04)
- LLM Review: Enhancing Creative Writing via Blind Peer Review Feedback (arXiv 2026-01)
- All Stories Are One Story: Emotional Arc Guided Procedural Game Level Generation (arXiv 2025-08)
- SuperWriter: Reflection-Driven Long-Form Generation with LLMs (arXiv 2025-06)
- Finding Flawed Fictions: Evaluating Reasoning via Plot Hole Detection (arXiv 2025-04)
- MLD-EA: Check and Complete Narrative Coherence via Emotions and Actions (arXiv 2024-12)
- Collective Critics for Creative Story Generation (EMNLP 2024-10)
- SWAG: Storytelling With Action Guidance (EMNLP Findings 2024-02)
- GROVE: Retrieval-augmented Complex Story Generation with a Forest of Evidence (EMNLP Findings 2023-10)
- EIPE-text: Evaluation-Guided Iterative Plan Extraction (arXiv 2023-10)
- Branch-Solve-Merge Improves LLM Evaluation and Generation (arXiv 2023-10)
- Re3: Generating Longer Stories With Recursive Reprompting and Revision (arXiv 2022-10)
- Model Criticism for Long-Form Text Generation (arXiv 2022-10)
H. Evaluation / Benchmarks / Metrics（75，仅列文字写作强相关代表）
- Illusions of the Gold Standard: Analysis of Human Evaluation Protocols for Long-form Text (arXiv 2026-06)
- Benchmarking LLM-as-a-Judge for Long-Form Output Evaluation (arXiv 2026-06) — code: cjj826/LongJudgeBench
- Narrative Flattening: How Post-Training Compresses Variation in LLM Fiction (arXiv 2026-05)
- Elias in the Lighthouse, Again? Diagnosing Low Diversity in LLM Stories (arXiv 2026-05)
- SAGE: Hierarchical LLM-Based Literary Evaluation via Ontology-Grounded Dimensions (arXiv 2026-05)
- StoryAlign: Evaluating and Training Reward Models for Story Generation (arXiv 2026-05)
- Spoiler Alert: Narrative Forecasting as a Metric for Tension (arXiv 2026-04)
- Lost in Stories: Consistency Bugs in Long Story Generation (arXiv 2026-03) ⭐ [[01_lost_in_stories]] — code: Picrew/ConStory-Bench
- Creative Convergence or Imitation? Genre-Specific Homogeneity in LLM Chinese Literature (arXiv 2026-03)
- STAGE: A Full-Screenplay Benchmark for Reasoning over Evolving Stories (arXiv 2026-01)
- LitBench: Benchmark and Dataset for Reliable Evaluation of Creative Writing (arXiv 2025-07)
- WritingBench: A Comprehensive Benchmark for Generative Writing (arXiv 2025-03)
- LongEval: Analysis of Long-Text Generation Through a Plan-based Paradigm (arXiv 2025-02) — code: Wusiwei0410/LongEval
- HelloBench: Evaluating Long Text Generation Capabilities (arXiv 2024-09) — code: Quehry/HelloBench
- Are LLMs Capable of Generating Human-Level Narratives? (EMNLP 2024-07)
- Art or Artifice? LLMs and the False Promise of Creativity (CHI 2023-09)
- （其余 ~59 篇涉及视频/角色扮演/文化偏见/多模态评估等，略）
I. Datasets / Surveys / Resources（32，仅列强相关）
- Narrative Theory-Driven LLM Methods for Automatic Story Generation and Understanding: A Survey (arXiv 2026-02) ← 综述，重点
- MUSE: Multi-agent Framework for Unconstrained Story Envisioning via Closed-Loop Cognitive Orchestration (arXiv 2026-02)
- StoryWriter: A Multi-Agent Framework for Long Story Generation (arXiv 2025-06)
- What Makes a Good Story and How Can We Measure It? A Survey of Story Evaluation (arXiv 2024-08)
- Weaver: Foundation Models for Creative Writing (arXiv 2024-01)
- Assisting in Writing Wikipedia-like Articles From Scratch (STORM) (arXiv 2024-02)
- CollabStory: Multi-LLM Collaborative Story Generation (NAACL Findings 2024-06) — code: saranya-venkatraman/CollabStory
- WritingPrompts: A Large-Scale Language Dataset for Story Generation (ACL 2020)
- （其余为漫画/角色数据集/文化叙事等，略）
J. Open-source Projects（10）
- 抓取时未解析出具体项目名，需到仓库对应章节查看。
面向"文字小说/长文写作"的精选清单（跨类挑选，⭐=已详读，链接直达笔记）
方法 · 长篇生成 - ⭐ DOME（大纲+记忆，NAACL 2024） - ⭐ Book-Writing Capability（2026-05，书籍级） - ⭐ Ex3（ACL 2024，抽取-扩写） - ⭐ STORYTELLER（2025-06，情节规划） - ⭐ POLARIS（2026-06，小模型写长文） - ⭐ LongWriter（2024-08，万字生成，工程必读）
方法 · 多智能体 - ⭐ Magnet（2026-07，角色驱动+世界状态） - ⭐ StoryWriter（2025-06）、⭐ StoryBox（2025-10） - ⭐ CogWriter（2025-02）、⭐ MUSE（2026-02，多模态）、⭐ Agents' Room（ICLR 2024，奠基）
结构 · 叙事技巧 - ⭐ Codified Foreshadowing-Payoff（2026-01，伏笔） - ⭐ Storyline Trees（2026-06，层级结构表示） - ⭐ Skeleton Coherence（2026-04） - ⭐ Spoiler Alert（2026-04，张力）、⭐ Suspenseful Iterative Planning（EACL 2024，悬念）
评估 · 质量 - ⭐ Lost in Stories（2026-03，一致性 bug） - ⭐ StoryAlign（2026-05，奖励模型） - ⭐ WritingBench（2025-03）、⭐ HelloBench（2024-09）（基准） - ⭐ Narrative Flattening（2026-05，后训练"叙事扁平化"） - ⭐ Story Evaluation Survey（2024-08，评估综述总纲）
中文场景 - ⭐ BiT-MCTS（2026-03，中文小说，双向 MCTS） - ⭐ Chinese Literature Homogeneity（2026-03，同质化诊断） - ⭐ CAT-LLM（2024-01，中文风格迁移）
尚未详读、值得补的 - 精修/自评：SuperWriter（2025-06）、R2-Write（2026-04）、Collective Critics（EMNLP 2024） - 综述/数据：Narrative Theory-Driven LLM Methods: A Survey（2026-02）、Weaver（创意写作基础模型）、STORM（Wikipedia 长文）