Vortex Previz

A spatial, agentic orchestration engine that turns narrative graphs into cinematic sequences using Gemini and Veo.

Comment

Inspiration

Traditional filmmaking tools are often linear and rigid. We wanted to build something that reflects the "Vortex" of creative ideation—a non-linear, spatial environment where ideas aren't just lines on a page, but nodes in a living narrative graph. Our goal was to turn the director into an orchestrator and the AI into a collaborative engine.

What it does

Vortex Previz is an agentic cinematic engine. Users spawn narrative nodes (Subject, Lighting, Action) in a 3D space. Our "Director Agent" (Gemini 3.1 Pro) then compiles these clusters into cinematic shot lists, orchestrating Google Veo for motion and Imagen for style, while providing real-time directorial commentary via the Web Speech API. It features "Timeline Sync," making 3D nodes pulse in harmony with the playing video.

How we built it

Built on a foundation of Google Cloud Run, the app leverages Python FastAPI and React. The core intelligence is powered by the Google GenAI SDK, which orchestrates Gemini 3.1 Pro for reasoning, Veo for video synthesis, and Imagen for high-fidelity keyframes. We used Google Cloud Build for a fully automated CI/CD pipeline and Firestore for real-time state persistence.

Challenges we ran into

The biggest hurdle was "Massive Multimodal Orchestration." Triggering high-quality video generation alongside image and text synthesis in parallel required a robust Server-Sent Events (SSE) architecture. Managing the long-running operations of Veo while keeping the UI responsive required complex state polling and a custom "Sub-Graph Compilation" engine.

Accomplishments that we're proud of

We are incredibly proud of the seamless "Timeline Sync" feature. Seeing a 3D node in the graph pulse with a neon aura the exact millisecond its corresponding shot plays in the video player is a massive win for UX. It creates a perfect loop between the creative "Brain" (the graph) and the "Body" (the film).

What we learned

We mastered the art of agentic orchestration. We learned how to use Gemini not just as a chatbot, but as a central director that controls multiple other AI models simultaneously. We also gained deep experience in deploying high-performance AI containers to Google Cloud Run for global scalability.

What's next for Vortex Previz

The next step is "Audio Genome Extraction." We want users to be able to drop in a musical track so Gemini can analyze the tempo and mood, automatically adjusting the cinematic pacing and "cuts" of the generated sequence to match the music's soul.

Built With

  • fastapi
  • framer-motion
  • gemini-3.1-pro
  • google-cloud-build
  • google-cloud-firestore
  • google-cloud-run
  • google-genai-sdk
  • google-imagen
  • google-veo
  • python
  • react
  • tailwind-css
  • three.js
  • typescript
  • vite
  • web-speech-api
Share this project:

Updates