Shot Caller

Shot Caller was built as a submission for the SensAI World Models Hackathon. The goal was to create a tool that helps filmmakers plan the production and flow of shooting a scene at scouted real-world locations, before ever setting foot on set.

The application uses the Marble Labs API to generate 3D scenes from any location with google street view. Three.js is then used to render the scene in the browser and place assets — actors, props, cameras — within it, giving directors and cinematographers an interactive pre-visualization environment.

An LLM pipeline parses the generated scene.json files and converts them into structured production documents, including shot lists and blocking notes, giving the production team immediately actionable outputs from a location scout, and a cost and budget breakdown.

Marble Labs API — used to reconstruct photorealistic 3D geometry and textures from street view imagery, creating an accurate spatial representation of scouted locations.

Three.js — handles real-time rendering in the browser and provides the interactive viewport for placing and previewing production assets within the generated scene.

LLM scene parsing — a language model reads the structured scene.json output and produces human-readable production documents, bridging the gap between 3D data and on-set workflow.