Generate full-length novels with AI! Harness the power of large language models to create engaging stories based on your prompts.
- Generate medium to full-length novels: Produce substantial stories with coherent narratives, suitable for novella or novel-length works.
- Easy setup and use: Get started quickly with minimal configuration required.
- Customizable prompts and models: Choose from existing prompts or create your own, and select from various language models.
- Automatic model downloading: The system can automatically download required models via Ollama if they aren't already available.
- Support for local models via Ollama: Run language models locally for full control and privacy.
- Cloud provider support (currently Google): Access high-performance computing resources for those without powerful GPUs.
- Flexible configuration options: Fine-tune the generation process through easily modifiable settings.
- Works across all operating systems
- Supoorts translation of the generated stories in all languages
- Resume interrupted generation runs from the last completed step
- Optional scene-by-scene generation pipeline for initial chapter drafts
- Optional final editing pass over the entire novel
- Optional scrubbing pass to remove potential AI artifacts (like leftover instructions)
- Includes a script (
Evaluate.py) for comparing the quality of two generated stories
Getting started with AI Story Generator is easy:
- Clone the repository
- Install Ollama for local model support
- Run the generator:
./Write.py -Prompt Prompts/YourChosenPrompt.txtThat's it! The system will automatically download any required models and start generating your story.
Optional steps:
- Modify prompts in
Writer/Prompts.pyor create your own - Configure the model selection in
Writer/Config.py
Not sure which models to use with your GPU? Check out our Model Recommendations page for suggestions based on different GPU capabilities. We provide a quick reference table to help you choose the right models for your hardware, ensuring optimal performance and quality for your story generation projects.
You can customize the models used for different parts of the story generation process in two ways:
You can override the default models by specifying them as command-line arguments:
./Write.py -Prompt Prompts/YourChosenPrompt.txt -InitialOutlineModel "ollama://llama3:70b" ...Available command-line arguments are stated in the Write.py file.
Key arguments include:
-Resume {path/to/run.state.json}: Resume a previous run from its state file.-InitialOutlineModel {model_string}: Model for initial outline generation and revision.-ChapterOutlineModel {model_string}: Model for per-chapter outline expansion or scene outline generation.-ChapterS1Model {model_string}: Model for chapter stage 1 (plot/scene writing).-ChapterS2Model {model_string}: Model for chapter stage 2 (character development).-ChapterS3Model {model_string}: Model for chapter stage 3 (dialogue).-FinalNovelEditorModel {model_string}: Model for the final novel-wide edit pass.-ChapterRevisionModel {model_string}: Model for revising chapters during the feedback loop.-RevisionModel {model_string}: Model for generating feedback/critique.-EvalModel {model_string}: Model for quality evaluation (true/false) and chapter counting.-InfoModel {model_string}: Model for extracting final story info (title, summary, tags).-ScrubModel {model_string}: Model for the final scrubbing pass.-CheckerModel {model_string}: Model for internal checks (e.g., scene JSON conversion).-TranslatorModel {model_string}: Model for translation tasks.-ExpandOutline: Expand the main outline into detailed per-chapter outlines before writing. (Enabled by default)-SceneGenerationPipeline: Use the scene-by-scene generation method for initial chapter drafts. (Enabled by default)-EnableFinalEditPass: Perform an additional editing pass on the entire novel after initial generation. (Enabled by default)-NoScrubChapters: Disable the final pass that cleans up potential AI artifacts. (Scrubbing enabled by default)-NoChapterRevision: Disable the feedback/revision loop during chapter generation. (Revisions enabled by default)-Translate {Language}: Translate the final story into the specified language (e.g., 'French').-TranslatePrompt {Language}: Translate the input prompt into the specified language before generation.
The model format is: {ModelProvider}://{ModelName}@{ModelHost}?parameter=value
- Default host is
127.0.0.1:11434(currently only affects ollama) - Default ModelProvider is
ollama - Supported providers:
ollama,google,openrouter - For
ollamawe support the passing of parameters (e.g.temperature) on a per model basis
Example:
./Write.py -Prompt Prompts/YourChosenPrompt.txt -InitialOutlineModel "google://gemini-1.5-pro" -ChapterOutlineModel "ollama://llama3:70b@192.168.1.100:11434" ...This flexibility allows you to experiment with different models for various parts of the story generation process, helping you find the optimal combination for your needs.
NOTE: If you're using a provider that needs an API key, please copy .env.example to .env and paste in your API keys there.
Edit the Writer/Config.py file to change the default models:
INITIAL_OUTLINE_WRITER_MODEL = "ollama://llama3:70b"
CHAPTER_OUTLINE_WRITER_MODEL = "ollama://gemma2:27b"
CHAPTER_WRITER_MODEL = "google://gemini-1.5-flash"
...The application automatically saves its progress to allow resuming interrupted runs. The state is saved after each major step, indicated by the last_completed_step value within the state file:
init: Saved right after initializing a new run or loading the state for resume, before any generation begins.outline: Saved after the main story outline is successfully generated.detect_chapters: Saved after the total number of chapters is detected.expand_chapters: Saved after expanding the main outline into per-chapter outlines (only if the-ExpandOutlinefeature is enabled).chapter_generation: Saved after each individual chapter is generated. This allows resuming mid-way through chapter writing.chapter_generation_complete: Saved once all chapters have been initially generated.post_processing: Saved just before starting the final editing, scrubbing, or translation steps.complete: Saved after the entire process finishes and the final output files are written.
The state is saved in a JSON file named run.state.json. By default, this file is located within the run-specific log directory:
Logs/Generation_YYYY-MM-DD_HH-MM-SS/run.state.json
Where YYYY-MM-DD_HH-MM-SS corresponds to the date and time the run was started.
To resume a run, use the -Resume command-line argument, providing the path to the specific run.state.json file you wish to continue from.
- Experiment with different local models via Ollama: Try out various language models to find the best fit for your storytelling needs.
- Test various model combinations for different story components: Mix and match models for outline generation, chapter writing, and revisions to optimize output quality.
- Generating decent-length stories: The system consistently produces narratives of substantial length, suitable for novella or novel-length works.
- Character consistency: AI models maintain coherent character traits and development throughout the generated stories.
- Interesting story outlines: The initial outline generation creates compelling story structures that serve as strong foundations for the full narratives.
- Reducing repetitive phrases: We're working on enhancing the language variety to create more natural-sounding prose.
- Improving chapter flow and connections: Efforts are ongoing to create smoother transitions between chapters and maintain narrative cohesion.
- Addressing pacing issues: Refinements are being made to ensure proper story pacing and focus on crucial plot points.
- Optimizing generation speed: We're continuously working on improving performance to reduce generation times without sacrificing quality.
We're excited to hear from you! Your feedback and contributions are crucial to improving the AI Story Generator. Here's how you can get involved:
-
🐛 Open Issues: Encountered a bug or have a feature request? Open an issue and let us know!
-
💡 Start Discussions: Have ideas or want to brainstorm? Start a discussion in our GitHub Discussions forum.
-
🔬 Experiment and Share: Try different model combinations and share your results. Your experiments can help improve the system for everyone!
-
🖊️ Submit Pull Requests: Ready to contribute code? We welcome pull requests for improvements and new features.
-
💬 Join our Discord: For real-time chat, support, and community engagement, join our Discord server.
Don't hesitate to reach out – your input is valuable, and we're here to help!
This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). This means that if you modify the code and use it to provide a service over a network, you must make your modified source code available to the users of that service. For more details, see the LICENSE file in the repository or visit https://www.gnu.org/licenses/agpl-3.0.en.html.
Join us in shaping the future of AI-assisted storytelling! 🖋️🤖