Dreamy: AI Fairy Tales

What it does

I have created a system that uses Google's latest AI model, Gemini, to generate video tales for kids. The AI model creates the story and image prompts, which are then passed to Mini DALL-E, powered by Google TPU Research Cloud on Hugging Face. The backend is built with Python 3, FastAPI, and Uvicorn. Once the images are created, the text of the tale is processed by /coqui-ai/TTS, installed on a machine hosted by Contabo, to generate a WAV/MP3 file that matches the video length. All elements are combined using MoviePy and returned to the user. Users interact via a POST endpoint that accepts prompts for one or multiple characters and returns the generated videos, which are stored as static files. A background job clears space in case of failures or if videos are older than 30 days. The frontend, built with React Native, features a simple UI with a video player, a random generation button on the home page, and a custom generation option on the create page, supporting up to two characters or a preselected one from a list. The app, designed for use by children or supervisors, is currently in testing and expected to be in production on the Play Store soon.