Skip to Content
DocsIntroduction

Introduction

AutoMovie lets your coding agent write a film’s screenplay and program its staging, motion and camera in TypeScript. The engine computes that code into the same frames every time. The result serves as a motion hint for video diffusion, or as a film rendered directly.

AutoMovie is a set of npm packages and a scaffold for a production, the project of one film. npx automovie create installs the scaffold: an AGENTS.md and CLAUDE.md, the skills that tell the agent how a film is researched, written, programmed, rendered and reviewed, the contracts its work answers, and an empty inspector page.

The screenplay

The screenplay is written through a chain of layers. A short runs brief → screenplay. A series runs settings → treatments → scripts → screenplay. Each layer cites the decisions of the layer above it by anchor, and the code cites the screenplay the same way.

The evidence graph, run by @wrtnlabs/evidence, turns every principle and decision about the world, the characters, the places and the era into an obligation. The layer that keeps an obligation answers it in writing: how it keeps it, or why it does not apply. So the screenplay cannot drift from its world, and the film follows the screenplay exactly. A reviewer in a separate context then reads each layer, and a review expires when the text it read changes. Authoring chain explains the layers and Evidence graph explains the obligations.

Figures and motion

A motion hint tells a video model where each person is, what they do, what they touch and when. The model supplies the surfaces. So every person in a production is a polygon man, plain primitives over the joints of a generated body, and every object is the simplest form that identifies it. Objects and sets can take more detail for a delivery; people stay the polygon man.

The accuracy goes into the performance. Each action brings the hands, feet and hips onto their targets, scaled to the body that performs it, with contacts and timing computed by the engine. Motion as code describes how a performance is written.

Three commissions

You decide where the work ends. The agent records that choice in the production’s PROJECT.md.

CommissionDone when
Screenplay onlyThe text is complete and handed off with its revision
Film-generation hintEvery main scene’s staging and motion is programmed and the whole video is assembled
Finished videoThe film is rendered directly in 3D, or generated by a video model from the hint, and accepted

Stills and isolated clips do not complete a film-generation hint. A commission can also end at a standalone asset, such as one motion or one image, which has no narrative.

Who does what

Each party owns a separate part of the work.

PartyOwns
YouThe commission, the decisions you keep or delegate, and watching and listening to the playback
Your coding agentResearch, the screenplay layers, the TypeScript, the stills it inspects and the records it keeps
AutoMovieThe scaffold, the contracts and checks, the motion vocabulary, the engine, capture and render

AutoMovie contains no LLM and no server. The agent is yours. The engine, the motion vocabulary and the inspector are libraries the production imports. An image or video model, when a production uses one, is reached through a provider you choose. The render runs on your machine through Chromium and FFmpeg.

Where to go next

Scene studies plays whole scenes made only of composed actions, and Motions lists the action vocabulary. The reference states the commands, the configuration and the APIs an agent writes against.

Last updated on