Skip to content
Work

Film pipeline · Claude skills · Packaged

The film pipeline is five Claude skills, one per production phase. The first locks the brief. Every later skill reads it, so references, live-action prompts and motion graphics all describe the same film.

A still from a demo film: a snow-covered castle on a cliff at blue hour, a lone rider crossing the frozen valley below.
A shot from our demo film, built on locked references.
Type
Claude skills
Status
Packaged
Built with
Claude skills · Seedance · Remotion

How it flows

You decide. Agents do.

Scroll and follow the signal. On the top lane a person makes the call. On the bottom lane the agents do the work.

You
Agents

01 / 05You

Brief.

Answer the interview until the film is decided.

A person makes this call.

YouAnswer the interview until the film is decided.

Film pipeline · 01 / 04

Lock the brief.

One document fixes every choice.

Loose notes snap into a brief document line by line: intention, look, camera, cast and deliverables, and the brief locks. A line drops from the brief to the references skill, which unfolds into three reference plates. The line then branches to three build skills side by side. Action turns a stack of prompt beats into a film frame, Mograph turns a motion sheet into a typographic frame, and Code motion turns lines of code into a bar chart frame. The three outputs and a reference plate travel into one film strip, and a line runs back up to the brief to show that everything came from it.

Why it exists

AI film output usually fails before the first prompt. Nobody decided what the film is, so every tool guesses, and each one guesses differently.

  • Without it: Generate, squint, re-rollWith it: One brief, decided once
  • Without it: Characters change between shotsWith it: References fix the faces and places
  • Without it: Every tool gets a different descriptionWith it: Every skill reads the same brief

What it does

  • Brief: interviews you until intention, look, camera, cast and deliverables are locked. No pixels

  • References: character turnarounds, location plates, style sheets and contact sheets, saved as reusable tokens

  • Action: live-action video prompts as scene context, beats and reusable segments, with a linter

  • Mograph: motion graphics from an AI video model, from one style-locked motion sheet

  • Code motion: a script becomes deterministic Remotion scenes, frame-exact on every re-render

  • Every skill has a prompt-only route that spends nothing

Where the AI sits

Four layers, you on top.

  1. 01

    You

    The intention, and the OK before anything is batched.

  2. 02

    Skills

    Five skills, one for each phase of a production.

  3. 03

    Models

    Image and video models, or code for motion graphics.

  4. 04

    Brief

    One document that everything else reads.

A closer look

A location plate: a frozen lake with standing stones and mountains in mist.

A location plate. Every shot set here carries this image.

A location plate. Every shot set here carries this image.
A character reference: a white dragon facing the camera on a dark studio background.

A cast reference, saved once and reused.

A cast reference, saved once and reused.
Code motion: a scene from our film How Much To Charge. Every frame is rendered from code.
One person splits into five specialists.
Scenes are timed to the voice-over, one line per scene.
Interface scenes are drawn as neutral glass UI, not screenshots.

Your version

Want one built?

Bring the workflow you have today. We will show you what this looks like for your team.

Next project

Sponsorflow

The sponsorship desk for a studio running several shows against a limited number of sponsor slots per month.