Articles in. Video out. Nobody babysitting it.

01Why

• The volume problem

  A content library grows faster than any team can film it. Turning an article into a video by hand costs a brief, a script, a voice, an edit and a review, and none of that scales with the number of articles.

• The wrong people on it

  Left alone, volume work eats the calendar of the people hired to have ideas. The expensive hours go to the cheapest part of the job. The team should be pushing the high concept work; the volume tier should run itself.

• SEO out, GEO in

  Search used to answer with links, so everyone wrote articles. Answer engines quote and summarize instead, and they surface video for more and more questions. GEO, generative engine optimization, is the craft of raising the odds that an AI answer mentions you and links back.

The twist is that the engines do not watch the film. They read everything wrapped around it.

Each film restates a ranking article’s claims in a second modality. The first audience is the scraper; the second is the person who stays for the answer. The script is the real product. The film is how it travels.

02The pipeline

One run, seven stations. A person appears exactly once, and at the station where a person is actually worth something.

1. 01Fetcha ranking article, the source of truth
2. 02Scriptbeats and voice over, written to the article's claims
3. 03Modulesshots filled from a library of house motion modules
4. 04Assemblean edit cut to film grammar
5. 05Renderheadless After Effects
6. 06QC gateslint, review and motion checks, before any person looks
7. 07Reviewa human at last: time-coded notes on the frame

07 → 02  Notes feed straight back in. The model reads, thinks, re-runs, and the cut comes back changed.

03How it is built

Two decisions carry the whole thing. What the pipeline knows is held in a graph rather than stuffed into a prompt, and every step it can take is a tool it reaches through one protocol rather than a chain of one-off integrations.

• A knowledge graph, not a prompt

  The pipeline reads from a graph of what the brand knows: products, claims, the language legal has already cleared, how one article relates to the next. A model asked to improvise from a page of prose invents. A model handed the relationships stays inside them.

• Taught the house style

  Before the first frame, the model studied the house: brand guidelines distilled into the graph, a shelf of finished films measured as benchmarks, production project files torn down to their keyframes. Every round of human notes is folded back in, so the pipeline that runs today is better than the one that ran last month.

• MCP as the wiring

  Every step is a tool behind the Model Context Protocol rather than a bespoke integration. The orchestrator asks for a script, an asset, a render, and does not need to know what is on the other end. Swapping a model or a vendor is a change of server, not a rewrite.

• Standards carried in the pipe

  Brand rules, disclosure requirements and the review gate ride along as part of the run. Governance built into the pipeline is the only kind that survives contact with a deadline.

04The creative director loop

We play creative director. The model does the rest. Review happens in a browser: watch the cut, draw a box on the frame, type the note. The notes land as time-coded feedback the model reads on the next run, the same way notes reach an editor, minus the meeting.

The pipeline proposes. It does not publish. What arrives is a cut ready to be judged, which is a much better use of a creative director than a blank timeline.

05What a run returns

• The film

  Every element, shape, word, image, movement, keyframe and edit made by the model in After Effects and Runway. Nobody touches the timeline; the only human input is feedback. Scored with music today, voice over next.

• The transcript

  For the article page and for YouTube. This is the text an answer engine can actually quote, which makes it half the point of the film.

• The structured data

  VideoObject JSON-LD for the page the film sits on, so an engine does not have to guess what the video says or why it is there.

• An editable project

  A real After Effects file, every layer and keyframe live. If an editor ever needs to take over a cut, there is no black box to fight.

• Variations

  The same script with a different cast and footage, rendered as easily as the first pass. Volume is the whole point.

06What it changes

The honest comparison is the quote for doing this the old way: a vendor pipeline, a year plus of work, seven figures for a template library. This replaced that math with a pipeline the team steers itself.

The obvious win is time and cost per video, and that one is real. The one worth more is what the team does with the hours it gets back. Volume that used to be a queue becomes something the pipeline handles, and the producers, editors and motion designers go back to the work that actually needs a person to have an idea about it.

The same argument runs through the rest of the production org: AI earns its place when it expands what an in-house team can make and how fast, not when it replaces the judgment about what is worth making. How the team is built.