From a topic to a finished film

3DLabs wanted a documentary studio, not a script generator. A creator types a topic and gets back a finished film: researched story, scene plan, footage, narration, sound, and a package ready for their editor.

The bar was a film the creator would put their name on, not a reel of impressive but random AI clips.

Every stage feeds the next. A weak still becomes a weak clip, which becomes a mistimed sound cue three stages later. Fully automatic meant one bad decision early could ruin a film nobody could fix.

Renders are also slow and expensive, and the best model for each stage keeps changing.

A studio that moves through the film one stage at a time, Story → Images → Video → Audio → Export, with the creator approving each step before the next one spends money.

Every scene shows four takes side by side, labelled by shot type (establishing, intimate, detail, atmospheric). Pick one, regenerate the rest, and move on. Credits, job history and notifications sit in the sidebar, so the creator always knows what is running and what it costs.

Behind the studio sits a ten-stage pipeline run by a job orchestrator. Each stage talks to a model-routing layer rather than a specific model, and a registry adds or retires models without touching the pipeline.

Credits are reserved when a job starts, confirmed when it succeeds and released when it fails. Workspaces keep each team's projects, assets and spend apart, and audit logs record what every run used.

3DLabs ships films a creator steers scene by scene, on a stack that can adopt next month's best model without a rewrite.