Video Published in 3 Minutes. AI Did the Work.

Mar 30, 2026 · 3 min read

🎬 Watch it in action: YouTube Video

Recording a talk is easy. Editing it is painful. So I automated most of it.

What It Does

Drop a raw recording into a folder. Wait a few minutes. Get back:

  • Trimmed video (cuts the "is this thing on?" parts)
  • Auto-generated captions
  • Thumbnail with title
  • Description and tags

Then it uploads to YouTube. Draft mode, so I can review before publishing.

The Tech

Whisper for transcription. FFmpeg for video processing. Claude for generating titles and descriptions. YouTube API for upload.

All glued together with Python. Nothing fancy.

The Interesting Part

Detecting where to cut. Turns out silence detection + transcript analysis works pretty well.

If there's 5 seconds of silence followed by "okay so let's get started," that's probably the real beginning.

What It Can't Do

Complex edits. Picture-in-picture. Fancy transitions. For that you still need a human.

But for simple "here's my talk recording" videos? Good enough.

Time Saved

Before: 2-3 hours per video.

After: 10 minutes of review and minor tweaks.

That's dozens of hours per year I don't have to spend on video editing.

← Back to Blog