What is videoclaw?

Welcome! If you have ever thought "I wish I could just describe a video and have it made for me" — that is exactly what this tool does. And it does it in a way you can trust, watch, and undo. Let's start with the big picture, then you'll be making your first video in minutes.

The one-breath version

videoclaw turns a one-line idea into a finished MP4 video — one careful, visible step at a time.

You type something like "a 15-second ad for my coffee brand" into your terminal. The tool writes a short brief, plans the shots, generates the clips using a real AI video service (Veo, Seedance, Runway, or Dreamina), stitches them together with narration and music, lets a human approve the result, and marks it published.

You run it with one command: vclaw.

Two ways to use videoclaw — pick the one that suits you

1. Just talk to an AI agent (easiest). videoclaw is built to be driven by an agent host like Claude Code, Claude Desktop, or Codex. You describe the video you want in plain English and the agent runs everything for you, pausing for your approval. No commands to memorize. → Use it with Claude Code (and other agents)

2. Run the commands yourself. Prefer to be hands-on? Every step is a simple command, shown throughout this guide and in the cheat sheet.

Either way it's the same videoclaw underneath. This guide teaches the commands so you understand what's happening — but you may never need to type them.

A note on the commands you'll see

Throughout this guide, commands are written for people running videoclaw from the source code on their machine, like this:

node dist/cli/vclaw.js video providers

If you installed the published package instead, you can drop the long part and just type vclaw video providers. Same result. We'll only mention this once — everywhere else, mentally swap in vclaw if that's how you installed it.

Think of it like an assembly line

Picture a factory line for video. Your idea goes in one end; a finished MP4 comes out the other.

Along the way there are stations: write the brief, plan the storyboard, generate the clips, stitch them together, get a human's approval, publish. At every station, videoclaw writes down exactly what it did in a plain file you can open and read. Nothing happens in a black box.

The best part: if something looks wrong at any station, you can rewind to that station and try again. You never have to start the whole thing over.

Every project lives in its own folder on your computer (at projects/<name>/), and that folder — not the screen, not some cloud — is the real source of truth. Each step saves a readable .json file inside it.

The two things that make it different

Lots of tools let you type a prompt and get a video. Here is what sets videoclaw apart:

1. Nothing is hidden. Every stage leaves behind a plain, readable file. And if a video provider fails, videoclaw tells you loudly — it will never quietly swap in a different provider and pretend everything is fine. What you see is what really happened.

2. It has a "stop and check" button. In director mode, videoclaw refuses to spend real money on the render until a human has looked at the storyboard and said "yes, go." That approval is a deliberate gate — it stays shut until you set VIDEOCLAW_APPROVE_STORYBOARD=1.

Rehearse for free

Almost every command accepts --dry-run. That plans the entire job — showing you exactly what would happen — without spending a single credit. When in doubt, dry-run first.

Who is it for?

• Creators and marketers who want repeatable, reviewable AI videos — product ads, presenter explainers, social campaigns, music videos.
• Small teams juggling many video projects who need one dashboard showing what's stuck, what needs a look, and what's ready to ship.
• AI agents — everything videoclaw writes is machine-readable, so an automated assistant can drive the whole pipeline on its own.

You do not need to be a programmer. If you can copy and paste a command, you can use videoclaw.

The steps at a glance

Here is the whole assembly line in plain words. Each step is a command, and each one saves its work to a file you can read.

1. init — Create a fresh project folder to hold everything for this one video.
2. brief — Turn your one-line idea into a short written brief.
3. storyboard — Break the brief into scenes, and auto-build a "story bible" so characters, settings, and props stay consistent.
4. plan — Pick the AI provider and prepare the exact request to send. (Also called execution-plan.)
5. produce — Actually generate the video clips. (Add --dry-run to rehearse for free. Also called execute.)
6. assemble — Stitch the clips together with narration and music into one MP4, then quality-check it.
7. review — A human (or the browser Review UI) approves the result.
8. publish — Mark it done and hand it off.

Where do "readiness" and the approval gate fit?

Between storyboard and review there's a small runtime layer that does the actual generating: readiness (is everything in place?), plan/produce (build and run the request), and execute-status / execute-cancel (check on or stop a running job). In director mode, a preflight check and the storyboard approval gate sit right before anything expensive runs. You don't have to memorize this now — the how it works page walks through it gently.

A handy rule to remember for later: a video is only truly ready to hand off when its review report says verdict: "pass" and publishReady: true. No guessing — the file says so.

Next: Install & setup

That's the whole idea. You now know what videoclaw is, why it's built the way it is, and the path every video follows. Time to get it running.

Head to Install & setup to get videoclaw on your machine, then make your first video.

When you want to go deeper, these sibling pages are waiting: how it works, modes, characters, providers, review & publish, assemble, troubleshooting, and the cheat sheet.