Let ChatGPT & Claude edit your video

One prompt, no editing software.

How it works 🧠

This isn't a video-editing button. ChatGPT Work and Claude Cowork are agents: you describe a finished result, and they work out the steps, write and run code in their own sandbox, and hand you back a file.

So when you give one a video and say "transcribe it, follow the instructions I say in it, and cut it for Reels", it will:

1. 1Pull the audio out and transcribe it, with timestamps.
2. 2Read the transcript and find your spoken instructions ("cut this", "caption this part").
3. 3Make the cuts, add captions, resize to 9:16 using video tools it installs or runs itself.
4. 4Give you an mp4 to download.

What you need ✅

• A raw talking-head clip. Short (under 2–3 minutes is safest), clear audio, one speaker.
• Say your edit notes in the video. Pause, then say them clearly: "Cut the next ten seconds." "Add captions from here." "Zoom in on this line." The AI reads these from the transcript.
• For ChatGPT: a paid plan. Work mode is on Plus, Pro, Business, Enterprise and Edu — not Free or Go. The desktop app is the safest place to try this.
• For Claude: a paid plan. Cowork is on Pro, Max, Team and Enterprise. Use the desktop app so Claude can read the video straight from a folder on your computer. Note: on Pro, Fable 5.1 uses usage credits from the first message. Only Max includes it in the plan. Opus 5.5 is included on Pro and handles this fine.

Option A — ChatGPT Work 💬

1. 1Open the ChatGPT desktop app and switch to Work mode.
2. 2Click the + button and add your raw video. If the upload is refused, use Open a local folder instead and point Work at the folder your video is in.
3. 3Paste the prompt below and send.

I've given you a raw talking-head video for Instagram Reels.1. Extract the audio and transcribe it with timestamps.2. Read the transcript and find every editing instruction I say   out loud (things like "cut this part", "add captions here",   "zoom in on this"). List them with their timestamps and   show me before you edit.3. Make those edits. Also: remove silences longer than 1 second,   add bold centred captions, and output 1080x1920 at 30fps,   H.264 with AAC audio.4. Install any tools you need in your own environment. Don't   upload my video anywhere external.5. Give me the finished mp4 to download, plus the edit list you   applied.

Option B — Claude Cowork 🧩

1. 1Open the Claude desktop app and switch to Cowork (on some Pro and Max accounts, Chat and Cowork are now merged, so just start a task).
2. 2Click the + button and add your raw video, or connect the folder it's in.
3. 3Paste the same prompt and send.

I've given you a raw talking-head video for Instagram Reels.1. Extract the audio and transcribe it with timestamps.2. Read the transcript and find every editing instruction I say   out loud (things like "cut this part", "add captions here",   "zoom in on this"). List them with their timestamps and   show me before you edit.3. Make those edits. Also: remove silences longer than 1 second,   add bold centred captions, and output 1080x1920 at 30fps,   H.264 with AAC audio.4. Install any tools you need in your own environment. Don't   upload my video anywhere external.5. Save the finished mp4 next to the original, plus a text file   with the edit list you applied.

Why the prompt is written this way 🔑

• "Show me before you edit." Rendering is the slow part. Catching a misheard instruction in the transcript costs seconds. Catching it in the finished video costs a re-render.
• "Install any tools you need." This is the line that makes it work. It gives the agent permission to set up transcription and video tools in its own sandbox instead of telling you it can't.
• "Don't upload my video anywhere external." Some agents will reach for an online transcription service. This keeps your footage inside the sandbox.
• Give it the output spec. Without the size, frame rate and codec, you'll get a file Instagram may reject.

Follow-ups that work 🔁

The cut at 0:42 is too early. Move it to 0:45 and render again.

Captions are covering my face. Move them to 70% of the frame height.

Also give me a 1080x1350 version for the feed.

The transcript has "Kodex" — it's "Codex". Fix it in the captions and render again.