OpenShorts: The Open-Source AI Clip Generator That Self-Hosts for Free

By Prahlad Menon 3 min read

Short-form clipping tools are one of the more aggressively monetized corners of the AI boom — $49/month subscriptions, watermarks, minute caps. OpenShorts is the open-source answer: an MIT-licensed AI clip generator you can self-host with Docker for $0, or run in the cloud if you’d rather not babysit a GPU.

What it actually does

Feed it a long video — a podcast, webinar, livestream, interview, or just a YouTube link — and it returns vertical 9:16 clips ready for TikTok, Reels and YouTube Shorts. The pipeline is refreshingly concrete and built from solid open components:

  • faster-whisper transcribes with word-level timestamps
  • PySceneDetect finds scene boundaries
  • Google Gemini 3.1 Flash-Lite scores the transcript and scene data to pick the 3–15 strongest moments (15–60s each)
  • FFmpeg cuts each moment
  • MediaPipe face tracking (with a YOLOv8 fallback) reframes to 9:16, stabilized so the crop holds still instead of swinging

No manual scrubbing, nothing to configure per video.

The clever bit: two-person framing

Most auto-croppers ruin interview footage — they either shrink the wide shot into letterboxed mush or lock onto one face and lose the other. OpenShorts instead stacks both speakers vertically when two people are on camera, drops the captions on the seam where they cover nobody, and switches back to a face-tracked single-person crop when the cut goes to one speaker. The AI picks the layout per clip automatically.

Beyond cropping

It’s grown into a three-tools-in-one platform:

  • Word-level subtitles burned in via FFmpeg
  • AI voice dubbing in 30+ languages through ElevenLabs, preserving the speaker’s voice, then re-transcribed so subtitles match
  • Hook text overlays — AI-written titles for the critical first seconds
  • AI video effects — Gemini-generated FFmpeg filter chains for color grading and transitions
  • A free “YouTube Studio” — thumbnail generator, 10 viral title suggestions, auto descriptions with chapter timestamps
  • AI UGC generator — scripted, lip-synced avatar videos with AI actors
  • Direct publishing to TikTok, Reels and Shorts from the dashboard

Built for AI agents

The part that makes this interesting for the automation crowd: OpenShorts ships an MCP server, REST API, and a zero-dependency CLI (pip install openshorts). You can wire Claude, ChatGPT or n8n straight into it:

# Self-hosted (no key needed):
claude mcp add --transport http openshorts http://localhost:8000/mcp

That turns “clip this 2-hour podcast into ten shorts and write the titles” into a single agent instruction — or an always-on autopilot channel.

What it costs

  • Self-hosted · $0 — MIT, Docker, bring your own API keys. No watermark, no cap. On a typical CPU an 8-minute video takes ~5–8 minutes to process.
  • Cloud · from $12/month — same software, compute and keys covered. An NVIDIA GPU clips that same 8-minute video in ~50 seconds. Free tier: first video up to 60 min, then 20 min/month (watermarked).

Why it matters

The economics of short-form content tooling have been quietly extractive. OpenShorts collapses a $49/month SaaS into an MIT repo you can run privately — the source video never leaves your infrastructure — while still offering a cloud tier for people who just want GPU speed. The MCP integration is the forward-looking piece: clipping stops being a dashboard chore and becomes something your agent does on command.

Repo: github.com/mutonby/openshorts · Site: openshorts.app