How to make videos with Claude: 3 routes that work
November 5, 2026 · 4 min read

Claude can't render a video file on its own. Every current Claude model takes text and images in and gives text out, so the footage has to come from somewhere else.
That still leaves Claude doing most of the work. It can call a video model for you, write the shot list that model needs, or write the code that renders the video. Here are the three routes, step by step.
What Claude does and doesn't do
Anthropic's model docs list the same thing for every current model: text and image input, text output. There is no video model inside Claude.
What Claude can do is use tools. A tool is any outside service Claude is allowed to call, such as a video model behind an API. The standard way to plug one in is MCP (Model Context Protocol), an open format for giving an AI app new tools.
So the question is not "can Claude make video". It's which tool you hand it.
Route 1: connect a video model through MCP
This is the shortest path from a chat to a finished clip. You add a video service once, and Claude calls it when you ask for a video.
In the Claude app
- In Claude, open Customize > Connectors.
- Click + Add, then Add custom connector.
- Give it a name and paste the server URL. For the Visual Sandbox MCP server that's
https://visualsandbox.com/mcp. - Sign in when the browser asks. There is no key to paste.
The Free plan allows one custom connector. Paid plans allow more. On Team and Enterprise plans, an owner adds the connector first and members then connect to it.
Now ask in plain words: "Make a 5-second vertical clip of rain on a neon street at night." Video is slower than chat. Claude starts a job, gets back a job id, and checks on it until the file is ready, usually one to three minutes.
In Claude Code
If you work in the terminal, Claude Code connects to the same kind of server with one command:
claude mcp add --transport http visual-sandbox https://visualsandbox.com/mcp
Then type /mcp inside Claude Code, pick the server, and finish the sign-in in your browser.
There's also a Claude Code plugin that adds the server plus prompt guides for each model in one install:
claude plugin marketplace add visualsandbox/skills
claude plugin install visual-sandbox@visual-sandbox
The terminal route pays off when video is part of a bigger job. Claude can make the product clips, download them into your project, and put them on a landing page in the same session.
One habit worth keeping: every video run costs money. Tell Claude to price each clip first and wait for your yes. On Visual Sandbox that's the estimate_cost tool, called before generate. It stops a loop of silent retries from eating your balance.
Route 2: script and storyboard in Claude
Even with a connector, the prompt decides the result. This is where Claude is strongest, so use it as your writer and shot planner.
Ask for a shot list, one line per shot. Each line should name:
- the subject and the one action it does
- the camera, such as "slow push in" or "handheld, eye level"
- the light and the place
- the length and the aspect ratio, 9:16 for Reels and TikTok, 16:9 for YouTube
Then ask Claude to turn each line into a separate prompt for the model you plan to use. One prompt per shot beats one long prompt for a whole minute. Video models handle one clear action far better than a story.
For a recurring character, have Claude write one fixed description and paste it into every shot. Changing "a woman in a red coat" to "a lady in a crimson jacket" between shots is how faces and clothes drift.
Route 3: have Claude write the video as code
Some videos aren't footage at all. Title cards, animated charts, captions and product walkthroughs are design work, and code renders them exactly.
Remotion is a framework that turns React code into MP4 files. Claude Code writes the code, and you watch a live preview as it changes. Remotion's own setup is short:
npx create-video@latest --yes --blank my-video
cd my-video
npx remotion skills add
npm run dev
The skills step adds Remotion's official guides for coding agents, so Claude follows its conventions instead of guessing. Ask for changes in plain words, then render with npx remotion render.
Remotion is free for individuals, non-profits and companies of up to three employees. Larger for-profit companies need a paid company license.
The two approaches mix well. Generate the footage through a connector, then have Claude place those clips inside a Remotion composition with your text and music on top.
Which route to pick
- You want footage from a prompt: add a connector (route 1).
- Your clips look random: fix the shot list first (route 2).
- You need exact text, data or branding on screen: use Remotion (route 3).
- You're making a full ad: do all three. Script in Claude, generate the shots, assemble in code.
Claude won't hold the camera, but it can run the whole crew. Pick the route that matches the shot you need, and let it do the rest.


