Make videos inside Claude

Connect once, then just ask. Claude writes the script, generates the narration and visuals, and hands back a finished video — without you opening a timeline, a browser tab, or this app.

The two ways to use it: on its own, or entirely through Claude.

Claude asks before it spends

When you ask Claude for a video, it does not generate one straight away. It first calls get_video_options and shows you the choices — shape, where the pictures come from, AI still quality, motion, look, voice and length — with what each one costs.

This matters because the choices are not equally free. Stock footage and stock images cost nothing. AI stills spend your Pollen, one generation per scene. And High or Maximum still quality needs a Pollinations tier that authorises them — having a balance is not the same thing, and a request your tier refuses comes back as a payment error and quietly falls back to stock.

If you have already said what you want — “a vertical one with free stock footage” — or you tell Claude to just get on with it, it will use your answer and sensible defaults rather than asking twice.

Why do it from chat at all

Most AI video tools are a website you sit in front of. That is fine for one video. It stops being fine when video is a recurring job — a weekly explainer, a product update, fifty localised clips — because every one of them costs you a context switch.

Talking to Claude removes the app from the loop. You are already in a conversation about the thing the video is about; the video becomes one more sentence in it. And because Claude can call the tools directly, it can do the follow-up work too: check on the render, look at the scenes, drop the one that missed, and re-cut.

Connecting it — about a minute

  1. Open Connect to Claude and sign in.
  2. Copy the connector link it generates. Your key travels inside that link, so there is no separate password to paste.
  3. In Claude, add it as a custom connector — and set Authorization to “None”.
The one step people get wrong. Setting Authorization to anything other than “None” makes Claude expect a second credential that does not exist, and the connection fails with an unhelpful error. The key is already in the link.

What you can ask for

Once connected, these are ordinary sentences, not commands to memorise:

“Make me a 60-second video about workplace hydration, in French.”
“Is that video done yet?”
“Show me the scenes.”
“Drop scene 3 — it doesn’t match the script — and re-cut it.”
“Take the end card off.”

Behind those sentences are six tools: create a video, check its status, list its scenes, remove scenes, set the end card, and re-cut. The last four are the interesting ones — editing by conversation rather than by dragging blocks around.

Editing without regenerating

Every video is kept as separate scene clips, each carrying its own picture and its own narration. So removing a scene is a re-stitch of clips that already exist, not a new render. It takes seconds, costs nothing, and the removed scene is kept — ask for it back and the video is rebuilt with it.

Automating it properly

Chat is one entry point. The same pipeline runs unattended:

SchedulesGenerate daily or weekly, optionally publishing to your own YouTube channel as each one finishes.
SeriesPlan a multi-part run — “six parts, one a week”. Each episode knows it is part 3 of 6, so it builds on the last instead of repeating the introduction, and the schedule switches itself off after the final part.
REST APIPOST /api/videos with your API key, from a script, Make, Zapier or n8n. Same engine, same metering.
WebhooksGet a signed callback when a render finishes, so your own system can take it from there.

The point of all of it is the same: the video pipeline should run whether or not anyone is watching it.

What it costs

Stock footage and stock images are unlimited and free. AI stills, AI-video hybrid and full AI video run on Pollen you connect yourself, so you set your own ceiling with your own provider and we never sit between you and that bill. Nothing here charges per seat.

Being straight about limits. Publishing to TikTok and Facebook is built but not live — both platforms must review the app before posts are allowed, so those buttons say “coming soon” rather than pretending. YouTube publishing is available once you connect a channel, and uploads are unlisted by default so an unattended schedule cannot broadcast to your subscribers by accident.

Connect it to Claude →

More guides: vertical video for TikTok, Reels and Shorts · turning a script into a video.