The AI Film Rig

A dedicated tool-maker agent builds and maintains the genmedia CLI — six specialized commands that wrap Google’s generative media APIs into a production-ready pipeline. Teams focus on creative decisions while the toolset handles API orchestration, file naming, resolution enforcement, and audit logging.

Early pilots had teams building their own tools; the process matured into a shared, battle-tested suite that every crew relies on. A companion Python batch module (genmedia.py) enables programmatic orchestration for bulk operations — generating dozens of storyboard frames or video clips without shell-quoting headaches.

Every team follows the same structured playbook — from concept to final cut. Each step has defined roles, outputs, and quality gates that must pass before advancing.

Maintaining recognizable characters across dozens of AI-generated shots is one of the hardest problems in generative filmmaking. The pipeline uses a reference chain methodology: each character gets 4+ reference images synthesized from a detailed text profile, then composed into a single Composite Character Sheet (multiple angles in one image). This sheet is fed as a reference image into every subsequent generation to anchor the character’s appearance.

The same approach extends to recurring objects and settings — a distinctive building, vehicle, or prop gets its own reference sheet so it looks consistent across every shot it appears in. Veo supports up to 3 reference images per generation, so teams budget their references carefully: typically one character sheet, one setting reference, and one object anchor.

Every shot is tagged with a vocal classification that enforces alignment between what the viewer sees and hears. Characters appearing to speak must have a dialogue track; narrator voiceover must play over shots where no one appears to be talking.