<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This repository offers a sophisticated, multi-stage workflow for generating original short stories using Large Language Models (LLMs). It addresses the challenge of creating high-quality, diverse, and AI-detection-evading creative content through a unique "fusion distillation" technique. Aimed at writers and researchers, it provides a structured approach to guide LLMs, ensuring originality and adherence to genre conventions.

How It Works

The core innovation is "fusion distillation," which extracts fingerprints (character, plot, style, keywords) from multiple source texts within a genre. The intersection of these fingerprints forms a generalized "type" fingerprint, used to prompt an LLM. This aims to produce content original and resistant to AI detection by not being attributable to any single source. The workflow supports a full pipeline or breakdown into 12 distinct creative stages, from inspiration to expansion, allowing fine-grained control and chained context passing.

Quick Start & Requirements

• Primary install / run command: pip install -r requirements.txt.
• Non-default prerequisites and dependencies: jieba Chinese word segmentation library; standard Python libraries. Requires preparing a corpus of .txt novels (min. 5 per category) in corpus/ subdirectories across 9 predefined genres.
• Links: No external demo or documentation links provided beyond the README.

Highlighted Details

• Fusion Distillation: Generates a genre-specific "type" fingerprint from multiple source texts, enabling LLM generation that is original and resistant to AI detection.
• 12-Stage Creative Pipeline: Offers granular control over writing stages (ideation to expansion), with support for chained context passing between stages.
• Automated Quality Audits: Includes tools for verifying originality against source corpora and checking against predefined "hit" rules for popular fiction.
• Local Execution: Operates entirely locally, requiring no external API keys.

Maintenance & Community

The README mentions learning from the "LinuxDo community" but provides no details on maintainers, active development, or dedicated community channels.

Licensing & Compatibility

Released under the MIT License, generally permitting commercial use and integration into closed-source projects with minimal restrictions.

Limitations & Caveats

Output quality depends heavily on the chosen LLM and post-processing. While originality against source texts is claimed, the AI detection evasion mechanism's robustness is not empirically demonstrated. The system relies on specific corpus preparation and genre categorization.