Workbench guide

How to Make Short Videos With Claude and Free Tools

Describe your video in plain words and let Claude build it: scenes, voice, captions, music and the final file. What it can do, what it cannot, and the exact prompts we used for a real 45-second reel.

A clay smartphone standing on a desk, opened up to show a tiny film studio inside with lights, a camera, a microphone and a director's chair. A yellow film strip spills out onto the desk.
A complete short-video production, built from a plan and free tools. The Engineer Desk, AI-generated illustration
In this piece

Why it matters

You do not need editing skills to make a clean short video. You need a clear plan, the right free tools, and a careful final check.

This guide shows a simple way to make short videos, like a 45-second product reel, without traditional editing. You describe the video in plain words. Claude builds it. You watch it and ask for changes.

We used this exact workflow for a real 45-second vertical reel for Sociora. Everything below is what that project actually used. Where we did not try something, we say so.

What Claude can and cannot do

Read this first. It saves the most frustration.

Claude can Claude cannot
Build every scene: text, images, timing and movement Watch or listen to your video
Run the tools that make the voice, captions and final file Create video footage, music or voices by itself
Look at still pictures from your video and spot problems Tell whether the timing feels right
Write the script and the captions Promise that a tool's license fits your use

The important part: Claude checks still pictures of your video, not the video itself. You still need to watch it on a phone, with headphones, before you post it.

What you need

Tool Cost What it is for
Claude Code Needs a paid Claude plan or API credits Builds and runs the whole project for you
Remotion Free for individuals, nonprofits and companies of up to 3 people. Larger companies need a paid license Turns Claude's work into a video
Kokoro voice Free. Code is MIT, voice model is Apache 2.0 A narration voice that runs on your own computer
Whisper (optional) Free, MIT license Captions when a real person is speaking
Freesound (optional) Free, but each sound has its own license Sound effects

You do not have to install these one by one. Claude Code sets them up in Step 1. If you have never used Claude Code, read our beginner guide first.

How it works

  1. Plan: write a list of scenes, how long each one lasts, and what is said.
  2. Build: Claude creates each scene.
  3. Voice: a free AI voice reads your script.
  4. Captions: captions are made from the voice's real timing.
  5. Sound: music and effects are added and balanced.
  6. Export: the finished video is saved as an MP4 file.
  7. Review: Claude checks still pictures, then you watch it yourself.

Step 1: Plan your video and set up the project

Write your plan as a simple list. For each scene: the time it starts and ends, what we see, and what the voice says. Keep it short. A 45-second video is about six to eight scenes.

Then make an empty folder, open Claude Code in it, and give it this prompt. Read what it plans to install before you say yes.

PromptSet up a short video project
Set up a new short-video project in this folder. Do not touch anything outside it.

The video: [LENGTH, FOR EXAMPLE: 45 seconds], [SHAPE, FOR EXAMPLE: vertical 1080 x 1920 for Reels and TikTok], 30 frames per second.

Please:
1. Set up Remotion. Tell me which Remotion license applies to me and link to the license page. I am [AN INDIVIDUAL / A COMPANY OF N PEOPLE].
2. Save my scene plan below in a file you can build from.
3. Set up the free Kokoro voice so it runs on this computer, and save its license.
4. Give me simple commands for: making the voice, making captions, mixing the sound, exporting the video, and making a contact sheet of still frames for review.
5. Do not upload anything, and do not use any paid service.

When you finish, explain in plain words what you set up, and list every license I should read.

MY SCENE PLAN
[SCENE 1: START-END TIME, WHAT WE SEE, WHAT IS SAID]
[SCENE 2: ...]

Step 2: Build one scene at a time

Ask for one scene at a time, not the whole video. Tell Claude what the scene must show and what must not move. Use your real material, such as your product, logo and fonts. Our reel used the product's real screens, not stock footage.

Step 3: Add the voice

The Kokoro voice reads your script on your own computer, so the script is never uploaded. It also records exactly when each word is spoken, which makes accurate captions possible.

Listen for two things:

  • Names. Our brand name was read wrong until we told the voice how to say it.
  • Honesty. The voice is AI. Do not present it as a real person, and never copy someone's real voice.

Step 4: Make captions that match the voice

Do not guess caption timing. Ask Claude to make captions from the voice's real word timing, in short groups of about four words. If a real person is speaking instead, [5] can write timed captions from the recording. Correct them by hand.

If the captions are part of the picture, turn off the app's automatic captions when you post, or viewers will see the words twice.

Step 5: Add music and sound effects

For our reel, Claude made the music and effects with a small program, so there was no license to worry about. If you download sounds instead, check each one. On [7], every sound has its own license, and some require credit.

Ask Claude to keep the voice clearly louder than the music. Our final mix measured about -16.7 LUFS, a common loudness for phone videos, with no distortion.

Step 6: Export the video

Ask Claude to export the final MP4, then check the real file, not just the settings.

PromptExport and check the video
Export the final video. Then check the actual MP4 file, not the project settings:
1. Tell me its size, frame rate and length, and whether it has sound.
2. Tell me how loud it is and whether any sound is distorted.
3. Make one contact sheet with a still frame every [N] seconds, and look at it. List anything cut off, overlapping, misspelled or hard to read.
Do not tell me it is finished until you have looked at the contact sheet.

Our final file was a standard MP4 (H.264 video, AAC audio) at 1080 x 1920 and 30 frames per second, which every major app accepts.

Step 7: Watch it yourself

Claude's still-picture check catches layout problems. It cannot tell if a joke lands late or a voice sounds odd. Watch the final file on your phone, with headphones, at least twice before you post.

Common mistakes

  • Expecting Claude to cut your existing footage. This workflow builds a video from scratch. To trim clips you already filmed, a regular editor such as DaVinci Resolve is often faster. We did not test Resolve for this guide.
  • Building everything at once. Work scene by scene, and check stills as you go.
  • Skipping the license check. Remotion, fonts, sounds and voices all have their own rules. Keep a folder of licenses.
  • Trusting the settings. Always check the exported file itself.
  • Showing what your product cannot do. Only show real features. Label sample data as sample data.
  • FFmpeg licensing. The tool that writes the final file is free, but its [6] matters if you ship it inside your own product. It does not matter when you simply use it.

Frequently asked questions

Can Claude edit videos?

Not like a normal video editor. Claude cannot watch video. In this workflow, it builds the video from a plan using free tools, then checks still pictures of the result.

Do I need to know how to code?

No, but it helps to be comfortable with Claude Code. Claude writes and runs everything. Your job is to give a clear plan, approve each step, and judge the result.

Is this workflow free?

The video tools are free for individuals and small teams. Claude Code itself needs a paid Claude plan or API credits.

Can I use this for YouTube Shorts, Reels and TikTok?

Yes. Export at 1080 x 1920 vertical, which all three accept.

Tools and resources

  • Remotion documentation How videos are built from code. remotion.dev
  • kokoro-onnx Runs the free Kokoro voice on your own computer, with word timing. github.com
  • Whisper Free speech-to-text for captions of real recordings. github.com
  • Freesound A library of sounds, each with its own license. freesound.org

Sources

  1. License and pricing, Remotion (remotion.dev)
  2. Remotion documentation, Remotion (remotion.dev)
  3. kokoro-onnx, GitHub (github.com)
  4. Kokoro, GitHub (github.com)
  5. Whisper, OpenAI on GitHub (github.com)
  6. FFmpeg License and Legal Considerations, FFmpeg (ffmpeg.org)
  7. Freesound FAQ, Freesound (freesound.org)
  8. Claude pricing, Anthropic (claude.com)
  • Video
  • Claude Code
  • Captions

Spot an error? Tell us. See how we handle sources and corrections in our editorial standards.

Found this useful? Get The Weekly Desk, one concise email a week.