Skip to content
View original post on X: vladasanadev
Published
28/100

Solo founder edits a video by speaking commands to Claude

AIA solo developer built a Claude Code skill, reel-live-edit, that adds logos, subtitles, background removal, and a clone to a talking-head video from spoken commands. The effects are timed to the words spoken, and the footage itself is the creator's real recording, not AI-generated.
vladasanadevVerified on X
@vladasanadev
View on X

i told claude to edit my video while i was talking.

logos started floating, my background disappeared, and suddenly there were two of me.

no editing timeline. just words.

here's how you can do it too https://x.com/i/article/2112863992368566272

Video Editing: Build a Video Editing Studio Around Opus 5.5

I told Claude to put a ChatGPT logo on my right hand, and it appeared exactly when I said it.

Then I asked it to make the logos float, add subtitles, change my background, and create a clone of me.

All inside one 35-second video.

The interesting part? None of the footage is AI-generated. It's my actual recording, with graphics, effects, and sound added through code.

I built a custom Claude Code skill to make this work, and I'm sharing the entire setup so you can try it yourself.

Here's how.

1. What we're making

The idea is simple: record yourself giving editing commands, then have Claude make each effect appear at the exact moment you say it.

Here's what happens in my video:

- "ChatGPT or Claude?" The two logos appear above my hands.

- "Let's make them edit in real time." The video transforms into a live editing interface.

- "Put ChatGPT logo on my right hand." A 3D cube appears on my palm and follows its movement.

- "Make them float and fly off." The logos start floating before flying off-screen.

- "Let's make it harder." A punch-in and difficulty animation appear.

- "Put subtitles." Animated captions begin.

- "Overlay a clip from any Tarantino movie." A floating video appears behind me.

- "Clear the background." My room disappears.

- "Put a clone of me next to me." A second version of me enters the frame.

The whole edit is synchronized to the words I'm saying.

2. What you'll need

You don't need a professional editing setup, but you'll need a computer capable of running the tools.

Software:

- Claude Code with a paid Claude plan

- Python 3.12

- FFmpeg

- Google Chrome

Hardware:

- An Apple Silicon Mac is ideal

- A Windows PC with an NVIDIA GPU also works

- Around 6 GB of free storage

Footage:

- A 30 to 40-second talking-head video

- Landscape, 16:9

- 1080p or 4K at 30 fps

- An optional movie clip for the overlay effect

Initial setup takes around five minutes. The actual editing process takes approximately 20 minutes, mostly for rendering.

3. Record your video

First, write a short script containing the commands you want Claude to execute.

Here's the script I used:

VIDEO SCRIPT

Who's better at video editing, ChatGPT or Claude?

Let's make them edit this video in real time.

Put ChatGPT logo on my right hand.

Now put Claude logo on my left hand.

Now make them float and fly off to the sides.

Let's make it harder.

Put subtitles using these two fonts.

Overlay a clip from any Tarantino movie when I say "movie".

Let's try even harder.

Clear everything on the background and put a clone of me next to me.

So... who won?

Want to edit videos like this too? Comment EDIT and I'll send you my instructions.

A few filming tips:

Keep your camera at eye level, with your upper body visible.

Leave some empty space on both sides of the frame. When you mention your hands, raise your palms and hold them in position until the floating animation.

Try not to cover your face when you get to the background replacement and cloning commands.

And don't worry about recording everything in one perfect take. Jump cuts are fine.

You can also switch ChatGPT and Claude for other AI tools, or replace the movie reference.

4. Install the Claude skill

I packaged the editing workflow into a downloadable Claude Code skill called reel-live-edit.

It includes the instructions, editing engine, script, and prompts.

Download the skill:

https://vladasana.com/blog/edit-with-vlada-ai/reel-live-edit.zip

Extract the ZIP file and move the reel-live-edit folder into your Claude Code skills directory.

On Mac:

~/.claude/skills/reel-live-edit

On Windows:

%USERPROFILE%\.claude\skills\reel-live-edit

You'll also need the supporting tools.

On Mac, install FFmpeg and uv using Homebrew:

brew install ffmpeg uv

On Windows:

winget install Gyan.FFmpeg astral-sh.uv Google.Chrome

If you're not comfortable using the terminal, you can ask Claude Code to help with installation.

Once the skill is installed, open Claude Code and use this prompt:

SETUP PROMPT

Use the reel-live-edit skill. Run its setup and tell me when it prints READY.

If anything is missing (ffmpeg, Chrome, Python 3.12), tell me the exact command to install it.

Claude will prepare the editing environment and download the models it needs.

Wait for the READY confirmation before moving on.

5. Let Claude edit the video

Now comes the fun part.

Give Claude the location of your recorded video and describe which AI goes on which hand.

Here's the editing prompt:

EDITING PROMPT

Use the reel-live-edit skill to edit my take: <path/to/take.mov>

The two AIs are ChatGPT (my right hand) and Claude (my left hand).

For the "movie" command use <path/to/clip.mp4>, starting at <m> in that file.

Work in ~/live-edit/my-take.

Before the long steps, show me the plan: every command, the word it lands on and the time.

Then render it, open the QA sheet, check every beat yourself, and give me the MP4 path.

No music, no AI-generated footage. Keep the ending clean (my footage + captions only).

What happens next?

Claude transcribes your speech and identifies the timing of individual words.

It detects cuts in your recording, tracks your hand movements, separates you from the background, and prepares the effects.

Then it combines the graphics, effects, audio, and footage into a finished MP4.

You don't have to manually place every animation on a timeline.

6. Review the editing plan

Before rendering, Claude generates a timeline showing where each effect will appear.

For example:

- 7.18s: ChatGPT logo appears

- 10.28s: Claude logo appears

- 12.70s: Logos begin floating

- 15.10s: Difficulty animation

- 20.17s: Movie clip appears

- 26.40s: Clone appears

These are example timings from my recording. Yours will depend on how you speak.

Check this plan before allowing the final render.

If an effect is attached to the wrong word, just tell Claude.

For example:

"The clone should appear when I say 'clone', not when I say 'put'. Move it and show me a preview."

After rendering, Claude produces a QA sheet with frames from the different effects, so you can inspect the result.

7. Fix anything that looks wrong

The first render doesn't have to be perfect.

You can make specific changes through normal instructions.

If a logo appears too late:

"Move the ChatGPT cube three frames earlier, show me a still, and re-render."

If the movie clip is too small:

"Make the movie card about 1000 pixels wide."

If the captions don't look right:

"Split the caption into two lines and adjust the timing."

If the background replacement leaves artifacts:

"Turn on the window matte cleanup."

If the logos appear on the wrong hands:

"Swap the brands. My video is mirrored."

I recommend fixing one issue at a time. Claude can preview individual changes before running another full render.

8. Make the Claude vs. ChatGPT comparison

I also wanted to see what would happen if I gave two AI tools the exact same editing task.

For a fair comparison, both tools need:

- The same original recording

- The same editing instructions

- The same requested effects

- The same output resolution

Ask both tools to deliver a 1920 × 1080 video synchronized to your original audio.

Once both edits are ready, you can use the skill to create a split-screen comparison.

COMPARISON PROMPT

Use the reel-live-edit skill to stack two edits of the same take into a 9:16 head-to-head:

Top = <claude_edit.mp4> labelled "CLAUDE"

Bottom = <other_edit.mp4> labelled "CHATGPT"

Title "EDITING" until the end of my first sentence, then "WHO'S better?".

Use my original take's audio.

Show me a frame before you send the file.

That gives you a vertical comparison video ready for social media.

Using the original audio keeps the comparison focused on the visuals rather than the different sound effects each tool might generate.

Try it yourself

That's the complete workflow.

You film a short video, install the skill, give Claude your footage, review the plan, and let it render the effects.

The skill handles the technical side. You decide what happens in the video.

Get the free skill and the full 11-page PDF guide here:

https://vladasana.com/blog/edit-with-vlada-ai/Live-Edit-Guide.pdf

One last thing: use video clips you have permission to include, and remember that the logos are for comparison, not endorsement.

If you try this, tag me @vladasanadev on X. I'd love to see what you make.

Source details

vladasanadev · x.comPublished · added here

Original title: i told claude to edit my video while i was talking.