🧠 Short GenAI Stories 📕

Every day, at midnight (UTC), a new short story is generated! You can read the stories online:

➡️ https://short-ai-story.web.app/

This project is developed in Java, and uses the LangChain4j open source LLM orchestration framework. Stories are generated thanks to Gemini, and Imagen 3.

You can read more about the inception of this project in this article:

➡️ An AI agent to generate short sci-fi stories.

The agent workflow

The ExplicitStoryGeneratorAgent class is an explicit workflow agent, in the sense that the code is driving the planning and execution of the story generation.

The agent:

• calls Gemini to create a story, with a title, with 5 chapters with a title and content for each
• for each of the chapters' content, asks Gemini to create a specific image generation prompt
• invokes Imagen with the prompt to generate a few images (4 by default) for each chapter's content
• a self-reflection step requests Gemini to judge the resulting images of each chapter to know which image matches the story line best.

Setup and Usage

1. Prerequisites:

   • Java 21
   • Maven
   • Google Cloud Project with Vertex AI enabled
   • Cloud Run jobs runs the story generator agent
   • Cloud Scheduler schedules the generation of stories every day at midnight UTC
   • Firestore, the document database to store the stories in
   • Firebase for serving the static assets and for accessing the Firestore database
   • Set the following environment variables:
     • GCP_PROJECT_ID: your Google Cloud project ID
     • GCP_LOCATION: the location of the region you want to run your model in (eg. europe-west1, us-central1...)
     • GCP_VERTEXAI_ENDPOINT: the endpoint for the embedding model (eg. europe-west1-aiplatform.googleapis.com:443)
   • Just — a handy tool to run the most frequent commands to work on this project (building, serving, deploying...)
     • The justfile uses the (non-committed) .env file to access and reuse the environment variables above

2. Serve and deploy static assets:

To serve the static assets locally, and browse the website locally, run:

just serve

To deploy the static assets on Firebase:

just deploy

3. Build the story generator agent:

The story generator agent is written in Java with LangChain4j. You can build the project with Maven, and package the application as a JAR with dependencies:

just package

4. Building the container, creating a Cloud Run job, and scheduling it

Build via Cloud Build:

just build

Create a Cloud Run job (create it only once):

just create-job

Schedule the job (create it only once):

just create-scheduler

Contributing

See CONTRIBUTING.md for details.

No promise is made on accepting contributions, as I don't intend to maintain this project in the long run.

License

Apache 2.0; see LICENSE for details.

Disclaimer

This project is not an official Google project. It is not supported by Google and Google specifically disclaims all warranties as to its quality, merchantability, or fitness for a particular purpose.