sequitur_studios

Sequitur Studios is a generative film studio built on the Gemini Omni Flash video model. It models a real production studio: the crew roles and departments of a film production, working across production phases, each grounded in film-craft domain knowledge instead of vague prompts. The studio has two halves: a crew that decides and a grammar that renders anywhere. Roles are first-class objects; a Cinematographer, Gaffer, and KeyGrip each own a slice of the shot grammar and choose their own values, while a Director reconciles their proposals into a single, complete Shot and a dumb Engine dispatches a phase. Each role delegates reasoning to a swappable Judgment, and the human is the Producer, who reviews each phase at a gate (approve or revise) while the agent crew executes.

One model-agnostic grammar drives three swappable backends: video (Gemini Omni Flash), still image (Azure Foundry gpt-image-1), and voice (Azure AI Speech text-to-speech). Today the studio composes across all three phases in code — a plan, a shoot, and an assemble — driven by a real crew engine and bound to a live Azure DevOps production board. The grounding library spans every department the architecture models, with nine film-craft sources imported and abridged. The studio is evolving toward a dailies model: an interactive, phase-gated pipeline where each phase emits a Producer-reviewable deliverable and the human approves or revises that phase before spend flows on. Its data plane is built, and the plan phase already emits its treatment, poster, and cast.