Kuala Lumpur
← All use cases

中文

Sales & marketing · Claude Code

Turn one keynote take into a captioned video with a Claude Code agent

Idea + starter prompt, not run by us yet

For: Marketers and speakers who have one short raw take and want a polished cut

A marketer built a Claude Code agent that lives in one folder, then gave it one keynote take, brand files and music. It wrote a transcript with a timestamp on every word, cut the video with ffmpeg, made a web page visual for each sentence and rendered with HyperFrames. After each step it had to show proof, like a transcript or a still frame.

Steps

  1. Make a folder for the agent. Put in one keynote take, your brand files and a music track you hold the licence for.
  2. Ask Claude Code to transcribe with word timestamps and show you the transcript before it cuts anything.
  3. Have it cut with ffmpeg without clipping words. Play a few cuts yourself to check.
  4. Ask for one visual per sentence and look at a still frame before the full render.
  5. Render a 16:9 and a 9:16 version. The source says setup takes about an hour and heavy daily use may hit plan limits. Its later prompts sit behind a signup, so expect to write your own.

Starter prompt

Starter prompt. Change the [brackets] first.

Use only public material or a sanitized working copy. Use fictional client details. Do not read credential files or request secrets. If private client data appears, stop without quoting it.

You are my video editing agent in this folder. My keynote recording is [video file name], my brand files are in [brand folder] and my music is [music file].

Work on copies only and never change the originals. Step one: transcribe with word timestamps and show me the transcript, then wait for my OK. Step two: list the cuts you propose, making sure no word is clipped, and wait for my OK before cutting with ffmpeg.

After each step, show proof such as a transcript excerpt or a still frame. Then propose one motion graphic per sentence as a web page and show me one still frame before you render. Add captions from the approved transcript. Use my brand assets and licensed music, then show me a short preview for approval before the full render. Save the 16:9 and 9:16 versions in a new folder. Ask me before you install anything.

Original post: futureofmarketing.de · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.