Inspiration

As a parent of a toddler there are times where i simply don't know what to do with her. I take care of my daughter on the weekends and we would play all of the games that we have and there inevitable would come the moment when i go - "What else is there to do?"

And so i came up with this idea, to create an app that generates a story based on what my daughter wants the story to be about. And then creates a coloring book with illustrations for each of the scenes for her to color the pages.

What it does

PixieTales Lab app generates a story based on the parameters set by user and proceeds to generate a custom coloring book. Users can set the name of the character, choose an animal that the character will be, select setting, add a theme or a special message that they want the story to focus on. Stories it generates have an educational message to the kids, they are coherent and have matching coloring-book-style illustrations—all ready to download as a PDF and share.

How I built it

PixieTales Lab is a web app with a React frontend (deployed on Firebase Hosting) and a FastAPI backend (deployed on Google Cloud Run). The core of the project is a multi-agent system orchestrated with Google’s Agent Development Kit (ADK). It uses Google Cloud’s Vertex AI for story and image generation. Content safety and child-friendliness are priorities, with prompt engineering and agent design ensuring gentle, positive stories and images.

Challenges I ran into

Maintaining illustration coherence: As anybody who have tried to generate a comic book using AI knows - consistency between images across one story is a big issue. It took awhile to come up with a set of prompts and a flow that results in a somewhat coherent style, although still from time to time it produces an image that doesn't quite fit with all the others.

Frontend/Backend Integration: Ensuring seamless communication between the React frontend and the FastAPI backend was a bit of a challenge.

Accomplishments that i'm proud of

-Built a fully functional, end-to-end pipeline for generating illustrated children’s books as downloadable PDFs. -Designed a modular, agent-based architecture that’s easy to extend and maintain. -Deployed a scalable, cloud-native solution using Firebase Hosting and Cloud Run. -Created a safe, accessible, and joyful creative tool for parents to entertain their kids with.

What I learned

How to use Google ADK. An awesome architecture. Very intuitive ! How use Vertex AI APIs and Google Cloud in general. The power of prompt engineering for both text and image generation to achieve coherent, high-quality results. Especially for images!

Built With

  • docker
  • fastapi
  • firebase-hosting
  • google-agent-development-kit-(adk)
  • google-cloud-run
  • google-vertex-ai
  • material-ui
  • node.js
  • python-3.9+
  • react
  • reportlab
  • typescript
Share this project:

Updates