Open Source · Beta

Turn your VODs into highlight clips automatically.

Paste a Twitch or YouTube VOD. Auto-Clipper watches every frame, finds the kills, combat, and explosions, and hands you ready-to-post clips — horizontal or vertical.

Free & open source No account needed Runs locally on your PC
Features

Everything you need, nothing you don't.

Computer vision built for gaming. Works offline. Outputs clips you can drag straight into OBS, Premiere, or TikTok.

Game-tuned computer vision

Frame-by-frame HSV color detection, motion deltas, scene changes, audio spikes, and optional YOLO object detection — all scored together per game profile.

HSV Color Motion Audio Scene Cuts YOLO

AI Mode

Optional xAI Grok Vision pass understands what is happening — not just colors and motion.

TikTok / Shorts export

One click converts any clip to vertical 9:16 with picker layouts — webcam, no-cam, or custom frame.

Local VOD library

Downloaded VODs stay on your machine. Re-analyze with a different profile in seconds — no re-download.

Preview & trim

Every clip gets a preview player with ±1s / ±5s trim buttons and instant re-cut. No external editor needed.

Built for real games

Per-game detection profiles — kill-feed colors, damage vignettes, explosion flashes, UI zones. Adding a game = adding a profile.

Runs on your PC, not the cloud

Your VODs never leave your computer. No uploads, no accounts, no data brokers. Just Python, FFmpeg, and your GPU.

New · Voice-triggered clipping

Say "clip it." We caught it.

A 30-second rolling buffer sits behind your live stream. The moment you yell "clip that" — we snip the last 30s, drop it in your library, and let you trim/extend it in the editor.

LISTENING · buffer full (30s)
  • "let's push this compound —"
  • "CLIP THAT" ✓ trigger
  • — saved clip: last 30s → editor

Pick your clipping mode

Switch per session. Combine for the most recall.

Trigger phrases: "clip it" "clip that" "clip this" "save that" "save clip" + add your own
How it works

Three steps. Zero editing.

From stream URL to downloadable highlight clips in minutes — not hours of manual scrubbing.

01

Paste a VOD link

Drop a Twitch or YouTube URL. Or upload a local file. Want just a segment? Set a start/end time — it only downloads that part.

02

Let it scan

Auto-Clipper samples frames, runs color & motion detectors, aligns audio peaks, and scores every second. A 1-hour VOD takes ~3–5 min.

03

Review & post

Get a grid of clips with thumbnails, timestamps, and confidence bars. Preview, trim, download, or one-click convert to TikTok vertical.

Performance

Fast. Measured, not marketed.

Real numbers from the 0.12 release on a 10-second smoke test and a 1-hour Twitch VOD. Your mileage depends on GPU / CPU — MPS and CUDA are autodetected.

Pixel-only mode
28.6×realtime
CPU only. 10s smoke test → 0.35s wall time. No weights, no GPU, no accounts.
YOLO on Apple Silicon MPS
~6–8×realtime
1-hr VOD → ~8 min end-to-end including download. Autodetects MPS / CUDA / CPU.
Clip It voice scan
~5–10×realtime
Whisper audio-only transcription (faster-whisper backend). Each trigger → 30s clip.
-100%
Temp-JPEG writes removed. Numpy frames go straight to YOLO.
-30%
Wall time in YOLO-only mode — pixel analysis skipped when not needed.
~60%
Decoder cycles saved — `cap.grab()` skips cvtColor + memcpy on non-sampled frames.
0
Network calls required for default detection. 100% on-device.
Who it's for

Built for the grind.

If you've ever finished a 4-hour stream and faced the scrub bar, this is for you.

Twitch streamers

End your stream. Drop the VOD link. Get a 30-clip reel before you finish your water. Say "clip that" mid-stream for instant highlights.

  • Twitch / YouTube VOD URLs
  • Auto-export TikTok 9:16 vertical
  • Rolling 30s buffer via voice

Content creators

Short-form pipelines that used to take hours: paste, wait, trim, post. Stack multiple VODs into a batch and grab the best moments across sessions.

  • Batch analyze via library
  • Cut + trim without re-encoding
  • Preview before you commit

Competitive / coaching

Pull every kill + death + close-call from a scrim. YOLO classes tag raiders, turrets, and bosses so you can filter your review to specific encounter types.

  • Per-entity clip filters
  • Confidence-ranked library
  • Share links to specific timestamps
Supported games

Tuned per game. More coming.

Each detection profile is hand-tuned for that game's UI, colors, and visual effects. Request a game on GitHub — or add one yourself.

Live

Arc Raiders

Muzzle flash, damage vignette, arc glow, kill flares, XP pops, scanner beams, and more.

  • Kill feed
  • Damage
  • Arc glow
  • Explosions
  • YOLO
Live

War Thunder

"Target Destroyed" messages, critical hits, bomb & rocket hits, vehicle fires, air combat.

  • Target destroyed
  • Crit hit
  • Fire
  • Bomb hit
Soon

Valorant · Apex · Fortnite

More profiles in progress. Want priority? File a GitHub issue with a sample VOD and we'll tune it next.

  • Vote on GitHub
Install

One command. Paste. Done.

Open Terminal, paste the line below, hit Enter. No warnings, no installers, no Gatekeeper dialogs.

The way to install · Mac & Linux

Paste one line in Terminal

Same pattern Homebrew uses. Files fetched via curl skip the macOS quarantine flag, so there's no "cannot verify" dialog — ever. The script clones the repo, installs Python + FFmpeg + deps, and launches the app.

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/bendawg2010/Auto-clipper/claude/twitch-clip-analyzer-MPT08/install-remote.sh)"
1 Open Terminal (⌘+Space → "Terminal")
2 Paste the line above, hit Enter
3 Browser opens at localhost:8080
On Windows? Open PowerShell and paste:
git clone https://github.com/bendawg2010/Auto-clipper.git; cd Auto-clipper; .\install.bat; .\run.bat
Git + Python 3.11 required — see python.org (check "Add to PATH").
FAQ

Questions, answered.

Is it really free?
Yes. Auto-Clipper is MIT-style open source and runs entirely on your machine. There are no accounts, no paywalls, and no data uploaded anywhere. The only optional paid feature is AI Mode, which uses your own xAI API key.
Do I need a GPU?
No. The default detector is CPU-friendly HSV color + motion analysis — a modern laptop handles a 1-hour VOD in ~3–5 minutes. If you enable optional YOLO object detection, a GPU makes it significantly faster.
How is this different from Twitch auto-clips?
Twitch highlights are based on chat activity and moderators. Auto-Clipper watches the actual gameplay pixels — it finds the moment a kill happens, not the moment viewers reacted.
What games are supported?
Arc Raiders and War Thunder today. Each game needs a detection profile tuned for its UI, colors, and effects — the engine is the same. Request a game on GitHub or submit a pull request with a profile of your own.
Why isn't it a website I can just open and use?
Auto-Clipper processes full VODs with FFmpeg and OpenCV — that's gigabytes of video and heavy CPU work per user. Hosting it for free isn't feasible. Running locally keeps it free, private, and fast.
Can I contribute?
Please do. Adding a new game profile is the easiest first contribution — open a PR on GitHub. Detection tuning and new export formats also welcome.

Your next highlight reel is one paste away.

Clone, install, open in your browser. Zero sign-ups. Zero uploads. Just clips.