Monofilm

An AI film studio with a human in charge

The AI proposes. You direct.

Monofilm reads your story, then works like a crew: it proposes the people, places and scenes, and you approve, fix or refuse every one before anything is built on it.

Open the studio View on GitHub

Built for the Meta Global AI Developer Hackathon. Track: Build for Your World, Indie Developer. Made with Meta's Muse Spark, Muse Image, SAM 3.1 and Muse Voice Transcribe.

The Monofilm studio on step one, Start with your story, with the sample story Ask First written in the page and a margin card on the right.

Step one of seven, with the demo story, Ask First: a deep-space station AI that must ask a human before it acts.

The problem

AI film tools are fast, and you cannot trust the result.

How most tools work

You type a prompt, wait, and get a finished clip. If a face drifts or a room changes between shots, you start again. Nobody decided anything, so nobody can say what to fix.

How Monofilm works

The AI makes one thing at a time and shows it. You decide. What you approve is kept, what you refuse is redone, and everything built on a change is marked out of date.

How it works

Three beats, repeated on every page.

  1. You tell it

    Write or paste a story, add notes, choose a look. Say what you want in your own words, or speak it.

  2. It makes something to review

    Muse Spark and Muse Image propose people, places, scenes and pictures. Nothing is final and nothing runs without you.

  3. You press one button to go on

    Approve it, or fix it first. Every version is kept, so you can always go back.

The seven steps

One guided path from a story to a finished pack.

The first five steps are decided once for the whole film. Scenes each get their own five pages. Colour tells you where you are.

  1. 01

    Start

    Write or paste a story. Add files.

  2. 02

    Understand

    See what the AI took from it. Correct it.

  3. 03

    Look

    Pick the style of the whole film.

  4. 04

    People

    Three faces each. Choose one, and a voice.

  5. 05

    Places

    Decided once, so rooms stay the same.

  6. 06

    Scenes

    Each scene: five pages, below.

  7. 07

    Ship

    Check the film in order. Download.

Inside each scene Brief Stage Pictures Sound Pack
The People step showing three candidate faces for the night engineer Ines Okafor, with a button under each to use that face, and a description of her voice below.

People. Three faces for each person. The one you choose anchors every later picture. The faces in this demo come from a stand-in image model.

The Brief page of scene one, The crack: place, people, the look inherited from the film, and an editable transcript with German lines and their English meaning.

Brief. Each scene is a small production: transcript, place, people, notes for the video maker.

The Understand step with story beat three highlighted in red and a margin card open beside it: what is wrong, say it in your own words, what the AI will do, the price in credits, and a Fix it button.

Click anything, fix it

Point at what is off. It changes only that.

Click anything the AI made and a margin card opens. Say what is wrong, in a few taps or your own words, or press the microphone and speak it. Pin it on the picture. Keep what is good.

  • See what it will touch and what it will cost before anything runs.
  • Press Fix it, or leave a note for later.
  • Everything is versioned. Change something and what was built on it is marked out of date, not silently redone.

The Stage

Place people and cameras before a single picture is made.

Drag each person, object and camera on a floor plan, or open a 3D editor. The studio checks that everyone is in frame and nothing is hiding behind a wall.

That layout feeds the picture model, and SAM 3.1 later checks each picture against it and says what drifted.

The Stage page: a floor plan seen from above with the AI ORCA and two cameras, their fields of view drawn as cones, a list of items to move, and the note that layout checks pass.

The floor plan after a few drags. One scene, two cameras, nothing out of frame.

The Pack page of a scene listing its files: brief, transcript in two formats, and for each clip a prompt, a list of what must not show, the start picture and clip settings.

The pack

Everything a video generator needs, in one .zip.

Choose what your generator accepts, a start and end frame, a start frame only, reference images or text only, and the pack is trimmed to fit. When you bring a clip back, the studio checks that it begins and ends on your pictures.

An honest note about videoMonofilm does not generate video. Meta has no video model, so the studio prepares everything for the generator of your choice and checks what comes back.

The Pictures page: your layout on the left, then candidate start pictures and end pictures with Use this and Check layout buttons. The candidates are plain stand-in cards in this demo.

Pictures. Candidates for each start and end picture, made from your layout. These are stand-in cards from the demo build.

The Ship step: a clapperboard header with film name, scene count, runtime and shape, then each scene with its progress bars and the film in order.

Ship. The film in order, what is still open, and one download.

Built with Meta's models

Four models, each with one job.

ModelWhat it does here
Muse SparkReads your story, drafts the people, places and scenes, and revises them after your feedback.
Muse ImageMakes faces, places, and start and end pictures from your stage layout.
SAM 3.1Checks a picture against the planned layout and says what drifted.
Muse Voice TranscribeTurns your spoken feedback into the note on the margin card.

For humans and for agents

The same studio, open to your AI agent.

Monofilm is also an MCP server. An agent can read a production, propose a change, or work a step, under the same rules as you: it proposes, a person approves, and every version is kept.

Built on Cloudflare Workers, D1 and R2 with plain vanilla JavaScript and no build step. The studio you click and the server your agent calls share one history.

Workers · D1 · R2 · MCP · no build step

Start with a story

Make the film your way, one decision at a time.

Bring a story, or use the sample, Ask First. You will be asked before anything is made.

Open the studio View on GitHub