Automating videos with the API

A REST API under /api/v1, authenticated with an API key you create in your account. It covers the same ground as the rest of the product:

• Videos — create a render from a URL and a brief, read its status, fetch the finished MP4.
• Sites — survey a domain, get a plan of which pages deserve a video, generate them.
• Characters and captures — the reusable assets a series shares.
• Shows — recurring episodes driven by a feed.
• Connections — register an MCP or HTTP source that a format can pull data from.

It is a Pro capability, included from $49 a month. The key is scoped to your account and your credit balance; there is no separate API pricing.

Rendering is asynchronous, and any integration has to be built around that.

1. 1Create. Post the URL, the brief, the aspect ratio and any budget cap. You get an id back immediately — the render has not started, let alone finished.
2. 2Poll. Ask for the status. You get the phase, the progress, and an estimate recalculated from how long this particular render has actually been running rather than an average.
3. 3Read the result. When it is done, you get the MP4 — and the warnings.
4. 4Check the warnings. A render can finish and still not be whole, and the status says done either way. An automated pipeline that ignores warnings is a pipeline that publishes films with holes in them.

Budget several minutes per render, more if it generates clips. Do not hold an HTTP request open waiting for one.

• Catalogue video. One video per product page, generated on publish, embedded on the page. The highest-value automation, because it runs against pages that already get traffic.
• Release notes as video. A film per release, generated from the changelog entry. create_animation needs no page at all, so the changelog text is enough.
• Refreshing stale videos. Re-render when the page changes. A demo video that matches the current product is worth more than a better-produced one that shows last quarter's interface.
• Per-locale variants. The same page, filmed with narration in each language you sell in.

Rank before you generate. The instinct with an API is to run it across everything; the result is a large credit bill and videos on pages nobody visits. Sort by traffic, take the top slice, measure, then extend.

1. 1Generating AI visuals by default. A pipeline with clips enabled across a catalogue is the fastest way to empty a balance — a house clip is 160 credits against a render's 40, and a named model can cost far more. Set ai_budget on every call, and default it to zero for bulk work.
2. 2Re-rendering unchanged pages on a schedule. Trigger on change, not on a cron. A nightly job over a static catalogue pays full price every night for identical output.
3. 3Ignoring the warnings. They are the only thing standing between an automated pipeline and publishing a film whose key scene silently fell back to a text card.

Renders run one at a time per account rather than all at once, so submitting two hundred jobs queues them rather than running two hundred browsers. Build for a queue: submit, record the id, poll with backoff.

A render that fails for a system reason is refunded automatically, so a retry after a genuine failure does not pay twice. A render that fails because the page could not be reached — a robot check, a sign-in wall, a dead URL — is refused before anything is charged, and it will keep being refused until the page changes. Retrying that one in a loop achieves nothing; check the reason and skip it.

For pages behind a login, register a capture session first from the VS Code or browser extension. It is scoped to that origin, expires within three days, and is a Scale-plan capability.