Pastiche Novel Generation Creating: Fan Fiction You Love in Your Favorite Author's Style

TL;DR

This work defines Pastiche Novel Generation, a task to imitate a target author’s style and narrative depth, and introduces WriterAgent, an LLM-based system guided by a curriculum of four tasks (language style learning, world-building, plot planning, stylish writing). WriterAgent uses WriterLoRA, a hierarchical, task-specific adaptation of LoRA with a shared foundation and specialized adapters, enabling efficient, context-aware learning across tasks. Evaluations on multilingual classics (Dream of the Red Chamber and Harry Potter) show WriterAgent outperforms strong baselines in both plot coherence and stylistic fidelity, aided by curriculum learning and careful preservation of foundational language style. The framework demonstrates practical potential for personalized storytelling and cross-lingual pastiche, while also addressing ethical considerations around authorship and originality by emphasizing transformation and transparency in AI-assisted writing.

Abstract

Great novels create immersive worlds with rich character arcs, well-structured plots, and nuanced writing styles. However, current novel generation methods often rely on brief, simplistic story outlines and generate details using plain, generic language. To bridge this gap, we introduce the task of Pastiche Novel Generation, which requires the generated novels to imitate the distinctive features of the original work, including understanding character profiles, predicting plausible plot developments, and writing concrete details using vivid, expressive language. To achieve this, we propose WriterAgent, a novel generation system designed to master the core aspects of literary pastiche. WriterAgent is trained through a curriculum learning paradigm, progressing from low-level stylistic mastery to high-level narrative coherence. Its key tasks include language style learning, character modeling, plot planning, and stylish writing, ensuring comprehensive narrative control. To support this, WriterAgent leverages the WriterLoRA framework, an extension of LoRA with hierarchical and cumulative task-specific modules, each specializing in a different narrative aspect. We evaluate WriterAgent on multilingual classics like Harry Potter and Dream of the Red Chamber, demonstrating its superiority over baselines in capturing the target author's settings, character dynamics, and writing style to produce coherent, faithful narratives.

Figures (12)

• Figure 1: Comparison of traditional story generation and personalized long-form novel generation, showcasing enhanced narrative depth, character development, and stylistic fidelity.
• Figure 2: The entire training process can be divided into two parts: the pretraining phase and the fine-tuning phase. During the fine-tuning phase, tasks are divided into three stages of increasing complexity: world-building learning, plot structure learning, and stylish writing learning. These stages are integrated using curriculum learning.
• Figure 3: Demonstration of the model's stepwise learning on Dream of the Red Chamber: from curriculum 1 to curriculum 1 & 2. The text colors indicate the corresponding problems. Chinese version is in Appendix \ref{['fig:step_chinese']}.
• Figure 4: Ablation performance of WriterAgent.
• Figure 5: Comparison of reference and generated texts from the baseline and our WriterAgent. We highlight the weaknesses of the baseline model and the strengths of our approach. Some linguistic nuances may be lost in translation; see Figure \ref{['fig:write_red_chinese']} in the Appendix (Chinese) for accuracy.