Katha Studio
Katha Studio is an interactive, real-time conversational film director built for the Google DeepMind Bangalore Hackathon.
Core trick
On cloud failure, shifts story generation to a local Gemma model while preserving the story state across the switch.
Description
Katha Studio is an interactive, real-time conversational film director built for the Google DeepMind Bangalore Hackathon. It orchestrates cloud AI and local models to create a resilient, high-velocity creative pipeline.
README excerpt
# Katha Studio 🎬✨ **Katha Studio** is an interactive, real-time conversational film director built for the **Google DeepMind Bangalore Hackathon**. It orchestrates cloud AI and local models to create a resilient, high-velocity creative pipeline. ## 🏆 Hackathon Tracks Targeted ### 1. Special Prize: Local-First Agents on Gemma **The Problem:** Cloud-based creative agents die completely when the internet connection drops, breaking the creative flow and violating strict data privacy requirements. **Our Solution:** Katha Studio implements a resilient **Offline Narrator Agent**. If the cloud connection fails (or API keys are invalid), the system automatically catches the error, maintains the story state, and seamlessly shifts the entire workload to a local **Gemma 4 (e2b)** model running via Ollama. It continues the "sense-decide-act" loop entirely on-device. ### 2. Problem Statement 3: High-Throughput Creative Workflows (NB2 Lite) **The Problem:** Traditional storyboard generation is slow and not load-bearing to the creative process. **Our Solution:** We use **Nano Banana 2 Lite** programmatically. As the user talks to the Director Agent, the system instantly generates high-fidelity 1K resolution scene compositions in the background. This real-time visual generation is load-bearing—the user relies on these instant visuals to make their next creative decision. ### 3. Problem Statement 4: Conversational Video (Omni Flash) **The Solution:** We chain the ultra-fast images from NB