Welcome to LLM StoryTeller, an interactive web application that leverages Large Language Models (LLMs) to help you craft captivating stories effortlessly. Whether you're a student, writer, or enthusiast, LLM StoryTeller provides a seamless experience to generate, refine, and download your unique narratives.
Note: The application interface is in Spanish, but don’t worry! We will walk you through each step in detail in this README. The interface is intuitive, and with the included explanations and screenshots, you’ll find it easy to follow and understand the workflow. Here, you can see the main dashboard of the application:
Now, you can experience LLM StoryTeller directly on Streamlit Cloud, thanks to the integration of free models provided by OpenRouter. This version showcases the functionality of the interface with a simplified and accessible experience. Unlike the original llm_storyteller.py script designed for local use with your own machine models, this online version (llm_storyteller_openrouter.py) is optimized for public interaction and can be accessed at the following link:
LLM StoryTeller on Streamlit Cloud
Explore the power of AI storytelling visually and intuitively. Try it out now and see how the interface seamlessly helps you craft your stories!
- 📖 About
- 🚀 Features
- 🔧 Installation
- 🛠️ Usage
- ⚙️ Configuration
- 💡 How It Works
- 📚 Story Examples
- 📄 License
- 📬 Contact
LLM StoryTeller is a Streamlit-based application designed to assist users in creating engaging stories through the power of AI. Instead of simply requesting a story from an LLM, the application guides the language models through a structured three-step process: generating a detailed story outline, crafting the narrative, and refining it for grammar and coherence. This approach ensures higher-quality results compared to a single-step prompt. Additionally, the application is highly customizable, allowing you to select different models, adjust creativity levels, and tailor the story's style and length to your preferences.
To ensure the application functions correctly, you need to have two OpenAI-compatible language models running locally on your machine, configured to serve requests through an endpoint at http://localhost:7860. These models should be compatible with OpenAI's API format to handle prompts effectively. If you don't have these models or prefer a different setup, you can modify the BASE_URL and AVAILABLE_MODELS sections in the code to point to other endpoints or adjust the model names to match your setup.
- Guided Multi-Step Process: Directs LLMs through outlining, writing, and reviewing to ensure higher-quality stories.
- Model Compatibility: Easily configure and run OpenAI-compatible models locally, such as Llama 1B or Qwen 1.5B.
- Customizable Story Parameters: Adjust creativity, choose narrative style, language, and story length.
- Intuitive Interface: Simple and responsive design with clear input fields for seamless interaction.
- Downloadable Stories: Save the final story as a text file with a single click.
- Flexible Configuration: Modify model endpoints and settings to fit your environment.
Follow these steps to set up LLM StoryTeller on your local machine:
- Python 3.7+: Ensure you have Python installed. Download Python
- Streamlit: Install Streamlit using pip.
-
Clone the Repository
git clone https://github.com/warc0s/llm-storyteller.git cd LLM-StoryTeller -
Create a Virtual Environment
python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate
-
Install Dependencies
pip install -r requirements.txt
-
Run the Application
streamlit run app.py
-
Access the App
Open your browser and navigate to
http://localhost:8501
Navigate to the sidebar to select models for each storytelling step, adjust the temperature for creativity, and set the language of your story.
Fill in the main character, secondary character, location, key action, desired length, and narrative style.
Click on the "✨ Generar Historia" button. The app will process your inputs through the selected models to create your story.
As the story is being generated, you will see real-time updates for each of the three internal steps:
- Outline Creation: The app generates a structured story framework.
- Story Writing: The detailed narrative is crafted based on the outline.
- Review and Refinement: Grammar, coherence, and overall quality are polished.
Each step's progress is displayed with clear messages, giving you transparency and confidence in the process.
Once generated, your story will be displayed in a formatted container. You can download the final version as a .txt file by clicking on the button "📩 Descargar Historia".
LLM StoryTeller offers various configuration options to tailor your storytelling experience:
The application currently supports Llama 1B and Qwen 1.5B, optimized by default for these smaller models running on CPUs. These options ensure compatibility and performance in a lightweight setup.
If you'd like to use other models or endpoints, you can customize the application by modifying the BASE_URL and AVAILABLE_MODELS variables in the llm_storyteller.py file. This allows you to adapt the app to your preferred models or configurations.
- Outline Model: Generates the story framework.
- Writing Model: Crafts the detailed narrative.
- Review Model: Enhances grammar and coherence.
Control the creativity of the generated content. Higher values yield more creative outputs, while lower values ensure consistency.
The Language field is a flexible text box where you can input any language of your choice without restrictions. This input is directly included in the prompt sent to the LLM, ensuring your story is crafted in the specified language.
Additionally, select the desired narrative Style from predefined options such as Mystery, Science Fiction, Romance, Fantasy, and Comedy to tailor the tone and feel of your story.
To summarize, here’s a clear overview of how LLM StoryTeller works, as this structured approach has proven to be the most effective for generating high-quality stories, especially when using smaller models with limited parameters:
-
Outline Generation: The application begins by creating a structured framework based on your inputs. This ensures a clear direction and logical flow for the story.
-
Story Writing: The framework is expanded into a detailed and engaging narrative, incorporating the chosen language, style, and length specifications.
-
Review and Refinement: Finally, the story is polished for grammatical accuracy, coherence, and overall quality, ensuring the end result is compelling and well-written.
This step-by-step process is optimized for smaller models, ensuring they can perform effectively and deliver results comparable to larger models. By guiding the LLM through these structured phases and incorporating prompt engineering techniques, LLM StoryTeller maximizes the potential of the models, ensuring they generate stories of superior quality compared to a single-step prompt.
You can explore examples of generated stories (using the cloud version) in the examples folder. This folder contains three stories, each showcasing the results from different models:
-
Fantasy Story: Created entirely (all three steps) using Gemma 9B.
- Demonstrates rich detail and world-building with consistent quality across all phases.
-
Science Fiction Story: Generated fully with Llama 8B.
- Highlights Llama’s ability to handle suspense and technical narratives effectively.
-
Comedy Story: Produced entirely with Mistral 7B.
- This example shows limitations in coherence and creativity, making it the least polished of the three.
Note: To achieve better results, I encourage you to experiment with combining different models for each of the three steps (outline, writing, and refinement). For instance, you might use Gemma for outlining, Llama for writing, and maybe Mistral for refinement, to play to each model’s strengths and create a more balanced final story.
This project is licensed under the MIT License.
If you encounter any issues or have suggestions to improve the application, feel free to reach out or open a pull request on GitHub. Your feedback is greatly appreciated!
- LinkedIn: Marcos Garcia
- GitHub: warc0s