CineFlow

Inspiration

Making a movie is exciting, but managing the logistics behind it is complete chaos.

Before a single camera starts rolling, assistant directors and production managers spend weeks manually breaking down scripts—counting props, tracking characters, figuring out location needs, and estimating costs for every single scene. One tiny script change can ruin the whole schedule.

We built CineFlow to solve this bottleneck. We wanted to build an autonomous AI crew member that turns raw scripts directly into actionable set logistics, saving indie filmmakers and studio teams hundreds of hours of tedious manual work.

What it does

CineFlow is an action-driven multi-agent system that automates script breakdown and set logistics.

• Script Breakdown Agent: Reads PDF scripts and automatically extracts scenes, characters, locations, props, and special equipment needs into clean, structured data.
• Logistics & Budget Agent: Calculates scene complexity, estimates resource costs, and checks inventory availability in real-time.
• Dispatch Agent: Generates instant daily call sheets, creates quick visual storyboards for scenes, and alerts crew members via instant notifications.

Instead of just chatting or answering questions, CineFlow takes action—updating databases, triggering external workflows, and creating ready-to-use production assets.

How we built it

We built CineFlow on Google Cloud using a multi-agent architecture designed for speed and security:

• Google Cloud Agent Builder: Used to orchestrate our sub-agents, manage playbooks, and ground responses using native Data Stores without complex manual RAG pipelines.
• Gemini & Document Processing: To parse messy PDF script files and extract structured JSON data for scenes and characters.
• Function Calling & Tools: Enabled agents to execute real-world tasks like querying databases, triggering Cloud Functions, and generating audio reads using Gemini TTS.
• Imagen 3: Automatically generates visual concept art and storyboard frames directly from scene descriptions.
• Secret Manager: Safely stores and manages all external API keys to keep the system enterprise-ready.

Challenges we ran into

• Handling Complex Script Formats: Screenplays have unique formatting (e.g., INT/EXT tags, parentheticals). Getting the agent to parse every scene accurately without missing key details took carefully tuned instructions and grounding.
• Agent State Handoff: Ensuring the Script Breakdown Agent passed clean, structured data to the Logistics Agent without losing context or introducing hallucinations.
• Tool Calling Reliability: Forcing agents to reliably trigger real database actions and external APIs instead of just responding with plain text.

Accomplishments that we're proud of

• Zero-Setup Prototyping: Built and tested our multi-agent network smoothly using the Google Cloud Console Emulator without getting stuck in local setup dependencies.
• Real Action over Chat: Successfully created an AI system that doesn't just summarize scripts, but actually outputs real call sheets, budget updates, and visual storyboards.
• Clean Multi-Agent Workflow: Built a seamless handoff process between planning, budget calculation, and action execution.

What we learned

• How to use Google Cloud Agent Builder to deploy production-ready agent playbooks without writing thousands of lines of boilerplate code.
• The power of Function Calling and MCP (Model Context Protocol) in making AI models interact safely with databases and live web services.
• How practical AI tools can make a real difference in streamlining traditionally manual industries like film production.

What's next for CineFlow

• Live Slack / WhatsApp Crew Bot: Allowing department heads (like costume, lighting, or props) to ask questions and update set status on the go via messaging apps.
• Automated Call Sheet Sign-offs: Letting actors and crew confirm their arrival times directly through automated text/email loops.
• Deeper IBM Enterprise Integration: Connecting directly with enterprise asset management systems to track high-value equipment rentals across multiple shooting locations.

Built With

• gemini-api
• google-cloud
• google-cloud-agent-builder
• ibm
• python