wfcz10086/AI-automatically-generates-novels

What it solves

This project provides an automated pipeline for long-form narrative content creation. It solves the common problem of "AI drift" in long stories, where characters forget their traits, plot holes emerge, and the writing style becomes repetitive or overly generic. It enables the creation of novels, screenplays, short drama scripts, and anime storyboards from a single initial premise while maintaining consistency across hundreds of thousands of words.

How it works

The system uses a pluggable architecture where content types, genres, model providers, and export formats are all defined as plugins. The core process follows a hierarchical generation chain: One-sentence premise $\rightarrow$ World Bible $\rightarrow$ Character Profiles $\rightarrow$ General Outline $\rightarrow$ Volume Outlines $\rightarrow$ Chapter Outlines $\rightarrow$ Full Text.

Key technical mechanisms include:

• Layered Memory: Instead of relying solely on a long context window, it uses a six-layer memory budget (L0-L5) combining immediate outlines, resident world data, recent summaries, mid-term summaries, FTS5-based retrieval, and hard constraints.
• Self-Improvement Loop: The system performs quality checks every few chapters, generating a "Writing Guide" for the book that is injected into subsequent prompts to correct recurring issues.
• Three-Tier Quality Gate: It audits content at the chapter level (clichés), window level (repetition), and book level (character arcs/pacing), automatically rewriting sections that fall below a quality threshold.
• Consistency Tracking: It maintains a structured state machine for character locations, physical status, and a registry of plot foreshadowing to ensure payoffs occur.
• Integrated Search: It uses a built-in search provider (via SearXNG) to research historical facts, laws, or technical details based on LLM-planned queries rather than simple keywords.

Who it’s for

• Authors and screenwriters who want to automate the bulk of long-form drafting.
• Content creators producing scripts for short dramas or anime.
• Developers looking for a modular framework to build specialized narrative AI agents.

Highlights

• Pluggable Architecture: Easily add new genres, content types, or export formats (txt, md, Fountain, SRT) without changing core code.
• Multi-Modal Output: One setting can generate a novel, a screenplay, or a storyboard.
• Robust Model Compatibility: Handles various "OpenAI-compatible" API quirks, including different reasoning field names and content shapes.
• Granular Control: Offers both fully automated and stage-by-stage modes, allowing users to edit any intermediate artifact (like the World Bible) before proceeding.
• Local Persistence: Uses SQLite and file-based storage for project state, allowing for breakpoint resumes and easy project portability.