README Copy Section

📖 Automatic Novel Generation Tool

中文文档 | English | 日本語 | Français | Sawcuengh

~~Currently I don't have much energy to maintain this project. The project brings no revenue, and with graduation approaching I have many other priorities. If time permits in the future I may consider a refactor using newer technologies. — 2025/09/24~~

>- ~~(2026/03/09): This project will be refactored soon, featuring modern implementations and fresh creative concepts.~~

> Update (2026/03/25): The refactored version has completed initial development (only the main framework is done, features are not yet available) and will be uploaded to the dev branch within a week. Subsequent development will also be synchronized on the branch.

<div align="center">

✨ Core Features ✨

| Module | Key Capabilities |
|-----------------------|-----------------------------------------|
| 🎨 Novel Setting Workshop | Worldbuilding / Character Design / Plot Blueprint |
| 📖 Intelligent Chapter Generation | Multi-stage generation to ensure plot coherence |
| 🧠 State Tracking System | Character development trajectory / Foreshadowing management |
| 🔍 Semantic Search Engine | Vector-based long-term context consistency |
| 📚 Knowledge Base Integration | Supports local document references |
| ✅ Automatic Proofreading | Detects plot contradictions and logical conflicts |
| 🖥 Visual Workbench | Full-process GUI for configuration / generation / proofreading |

</div>

A multifunctional novel generator built on large language models. Helps you efficiently create long-form stories with consistent settings and rigorous logic.

---

📑 Table of Contents Copy Chapter

---

🛠 Environment Preparation Copy Chapter

---

📥 Installation Copy Chapter

git clone https://github.com/YILING0013/AI_NovelGenerator

2. Install build tools (optional)
- If some packages fail to install, visit Visual Studio Build Tools to download and install C++ build tools required by some modules.
- By default the installer includes MSBuild only; make sure to select C++ Desktop Development from the workload list.

3. Install dependencies and run
- Open a terminal and change to the project directory:

cd AI_NovelGenerator

python -m venv .venv
       # if that doesn't work, try:
       # python3 -m venv .venv

# On Windows:
       .venv/Scripts/activate

# On Linux/Mac:
       source .venv/bin/activate

pip install -r requirements.txt

python main.py

If some dependencies are still missing, manually run:

pip install <package-name>

🗂 Project Structure Copy Chapter

novel-generator/
├── main.py                      # Entry file, runs the GUI
├── consistency_checker.py       # Consistency checks to prevent plot conflicts
|—— chapter_directory_parser.py  # Directory parsing
|—— embedding_adapters.py        # Embedding interface wrappers
|—— llm_adapters.py              # LLM interface wrappers
├── prompt_definitions.py        # AI prompt templates
├── utils.py                     # Utility functions and file operations
├── config_manager.py            # Configuration manager (API keys, base URL)
├── config.json                  # User configuration (optional)
├── novel_generator/             # Core chapter generation logic
├── ui/                          # Graphical user interface
└── vectorstore/                 # (Optional) Local vector DB storage

---

⚙️ Configuration Guide Copy Chapter

📌 Basic configuration (config.json) config.json)')" title="Copy section prompt for LLMs"> Copy Section

{
   "last_llm_config_name": "DeepSeek V4 Flash",
   "llm_configs": {
      "DeepSeek V4 Flash": {
         "api_key": "",
         "base_url": "https://api.deepseek.com",
         "interface_format": "DeepSeek",
         "model_name": "deepseek-v4-flash",
         "temperature": 0.7,
         "max_tokens": 8192,
         "timeout": 600
      },
      "DeepSeek V4 Pro": {
         "api_key": "",
         "base_url": "https://api.deepseek.com",
         "interface_format": "DeepSeek",
         "model_name": "deepseek-v4-pro",
         "temperature": 0.7,
         "max_tokens": 32768,
         "timeout": 600
      },
      "Gemini 3.5 Flash": {
         "api_key": "",
         "base_url": "https://generativelanguage.googleapis.com/v1beta",
         "interface_format": "Gemini",
         "model_name": "gemini-3.5-flash",
         "temperature": 0.7,
         "max_tokens": 32768,
         "timeout": 600
      }
   },
   "embedding_configs": {
      "OpenAI": {
         "api_key": "",
         "base_url": "https://api.openai.com/v1",
         "interface_format": "OpenAI",
         "model_name": "text-embedding-3-small",
         "retrieval_k": 4
      }
   },
   "choose_configs": {
      "architecture_llm": "Gemini 3.5 Flash",
      "chapter_outline_llm": "Gemini 3.5 Flash",
      "prompt_draft_llm": "DeepSeek V4 Flash",
      "final_chapter_llm": "DeepSeek V4 Pro",
      "consistency_review_llm": "DeepSeek V4 Flash"
   }
}

🔧 Explanation Copy Section

2. Embedding model configuration
- embedding_model_name: Embedding model name (e.g., text-embedding-3-small, gemini-embedding-2, or Ollama's nomic-embed-text)
- embedding_url: Service endpoint
- embedding_retrieval_k: Number of nearest neighbors to retrieve

3. Novel parameters
- topic: Core story theme
- genre: Genre
- num_chapters: Total number of chapters
- word_number: Target words per chapter
- filepath: Path to save generated files

---

🚀 Run Instructions Copy Chapter

Method 1 — Run with Python Copy Section

python main.py

Method 2 — Build an executable Copy Section

pip install pyinstaller
pyinstaller main.spec

---

📘 User Guide Copy Chapter

2. Click "Step1. Generate Settings"
- The system will generate, based on topic/genre/chapter count:
- Novel_setting.txt: Worldbuilding, characters, trigger points and foreshadowing.
- You can view or edit these settings after generation.

3. Click "Step2. Generate Directory"
- The system will use Novel_setting.txt to produce:
- Novel_directory.txt: Chapter titles and short prompts.
- You can review and modify chapter titles and descriptions.

4. Click "Step3. Generate Chapter Draft"
- Before generating a chapter you can:
- Set the chapter number (e.g., 1)
- Provide chapter-specific guidance in the "This chapter guidance" box
- When you generate a chapter the system will:
- Read prior settings, Novel_directory.txt, and finalized chapters
- Use vector retrieval to recall relevant context for coherence
- Produce an outline (outline_X.txt) and chapter text (chapter_X.txt)
- You can view and edit the draft in the editor pane.

5. Click "Step4. Finalize Current Chapter"
- The system will:
- Update the global summary (global_summary.txt)
- Update character states (character_state.txt)
- Update the vector store (so future chapters can use the latest info)
- Update major plot points (e.g., plot_arcs.txt)
- After finalizing you will see the finalized text in chapter_X.txt.

6. Consistency check (optional)
- Click the "[Optional] Consistency Proofread" button to scan the latest chapter for conflicts (character logic, plot contradictions, etc.).
- If conflicts are detected, detailed messages will appear in the log area.

7. Repeat steps 4–6 until all chapters are generated and finalized.

Vector retrieval tips:

1. Explicitly set the embedding interface and model name.

2. For local Ollama embeddings start the Ollama service first:

``bash

`

3. Clear the vectorstore` directory after switching embedding models.

4. For cloud embeddings ensure the API permissions are enabled.

---

❓ FAQ Copy Chapter

Q1: Expecting value: line 1 column 1 (char 0) Copy Section

This error usually indicates the API did not return valid JSON—sometimes an HTML error page or other unexpected content was returned.

Q2: HTTP/1.1 504 Gateway Timeout? Copy Section

Check the stability of the API endpoint and network connectivity.

Q3: How do I switch Embedding providers? Copy Section

Enter the new provider settings in the GUI fields for embedding configuration.

---

If you have further questions or feature requests, please open an issue on the project repository.

---