I typed a few plain-English sentences into an AI in my terminal. A while later I had a working system that turns a script into a finished video — the same system that produces every video on this channel. I haven't written a line of its code by hand since.
Here's the honest version of how it works, the real tools under the hood, and how to build your own. No code experience needed.
## First, the honest part
The hook is true — it started with a few sentences. But a system you'd actually ship isn't one magic prompt. It's a loop: the AI writes a piece, runs it, sees what breaks, fixes it, and you steer it over a handful of passes. The sentences start it. The loop finishes it.
That's good news, not bad. It means you don't need to know how to build any of it. You need to describe what you want clearly and keep the loop going.
What is Claude Code
Claude Code is an AI that lives in your terminal — not a chatbot you copy answers out of.
It reads your actual files, writes code, runs it, sees the errors, fixes them, and repeats until the job is done. You describe what you want in plain English. It does the building.
You don't need Python, JavaScript, or any language. You need to know what you want and roughly what tools should do it.
The real stack (this is what most explainers hide)
"3 sentences, no code" makes it sound like magic. It isn't. Claude Code is the orchestrator — but it wires together four real, public tools. Knowing them is the difference between a vague toy and the actual system:
| Layer | Tool | What it does |
|---|---|---|
| Orchestrator | Claude Code | Reads files, writes the pieces, runs + fixes them in a loop |
| Runtime | Node.js | Runs every script |
| Voice | ElevenLabs | Turns a text script into a voiceover — with per-word timing |
| Visuals | Remotion | Video built in React; each beat is a component, rendered to frames |
| Stitch | ffmpeg | Muxes the voiceover and the rendered video into the final MP4 |
When I asked for the system, I named these tools in the prompt. That's the trick. "Build me a video tool" gets you nothing. "Build me a pipeline using ElevenLabs, Remotion, and ffmpeg" gets you a real architecture.
How the pieces connect
The part that makes it look produced instead of slapped together: ElevenLabs returns the timing of every spoken word. That word timing becomes the clock. Every visual beat is timed to a word, so the visuals always land on the voice.
script.txt
-> ElevenLabs -> vo.mp3 + words.json (voiceover + per-word timing)
-> Remotion -> reads words.json, renders timed visual beats -> video.mp4
-> ffmpeg -> mux voiceover + video -> final.mp4
One command in, one finished video out.
The exact prompt I started with
I have an existing project here. I want to turn a voiceover script into a
finished video automatically.
Build me a complete video skill — text-to-speech, a visual engine, music
and SFX timing, and a rendering pipeline — all connected, so one command
takes a script.txt and produces a final MP4.
Use Node.js, ElevenLabs for the voiceover, Remotion for the visuals, and
ffmpeg to stitch it together. Read what's already in the folder first.
The last sentence does the heavy lifting. Naming the actual tools is what turns a wish into a buildable spec.
The full prompt set — including the follow-up prompts that did the real refinement — is in the starter pack above.
How to do this yourself
You don't need to be technical. You need three things:
- A computer with a terminal (every Mac, Windows, and Linux machine has one)
- Claude Code — a free CLI tool from Anthropic
- A clear description of what you want, including the tools it should use
Step-by-step:
- Install Claude Code: in your terminal, run
npm install -g @anthropic-ai/claude-code - Go to your project folder:
cd your-project-folder - Start it:
claude - Describe your project in a sentence, then make your request — name the tools
- Let it work. When something breaks, tell it what broke. Keep the loop going.
- Test the result on a real input, not a toy one
The mindset that matters: treat it like a junior developer you can hand a clear spec to. Not a search engine. Not a chatbot. A builder you steer.
What most people get wrong
They give vague requests and get vague results.
Wrong: "Make me a video tool."
Right: "Build a step that takes script.txt, calls ElevenLabs, and saves vo.mp3 plus a words.json of per-word timestamps to /out."
The more specific you are about inputs, outputs, the tools, and where files live, the better the result. You don't explain *how* to build it — you describe what success looks like. Naming ElevenLabs, Remotion, and ffmpeg in the request is exactly this in action.
Tools & links
| Tool | Why | Link |
|---|---|---|
| Claude Code | The AI that builds and fixes the system | claude.ai/code |
| Node.js | Runs Claude Code and every script | nodejs.org |
| ElevenLabs | Voiceover + the per-word timing that syncs everything | elevenlabs.io |
| Remotion | Build video in React — visuals as code | remotion.dev |
| ffmpeg | Stitches voiceover and video into the final MP4 | ffmpeg.org |
The starter pack at the top of this page has all of it wired into a working skeleton — the prompts, a minimal ElevenLabs + Remotion + ffmpeg build you can run, and a setup guide. Start there, then hand it to Claude Code and describe what to add next.
One step to take today
Pick one repetitive thing you do every week that takes more than 15 minutes.
Write three sentences: what it is, what goes in, what you want to come out — and which tools should do it.
Then open Claude Code and say it. You'll be surprised how far a clear ask goes.
