Skip to content

Repository files navigation

OpenClipper

Deploy with Clawnify

Turn a long video into short vertical clips. Add a podcast, tutorial or talk (hours long, up to 30 GB) and OpenClipper finds the strongest moments, cuts them on clean sentence boundaries, lays each one out for 9:16 and burns in captions. Every clip is yours to trim, retitle, drop or re-render.

Why

Clipping tools score clips with a "virality" number and meter you by the credit. The part that actually costs a team days is simpler and more honest: find the moments that stand on their own, start each on its hook, end it on a finished thought, and frame it for a phone. OpenClipper does that, shows you why it picked each moment, and leaves the final call to you.

Features

  • Long videos, any size: resumable uploads straight from the browser (up to 30 GB and 4 hours) or import from a link. The source never passes through the app, and a clip only reads its own seconds of it.
  • Whole-video judgment: the model reads the entire timed transcript at once, so the picks are the strongest in the video rather than the first good ones, and never two clips of the same idea. It returns fewer clips when the video doesn't have enough strong moments.
  • Clean cuts: every clip is snapped to where speech starts and stops, so nothing opens or closes mid-word.
  • Layout per shot: a person filling the frame gets a 9:16 crop around them; a screen recording or slides stays whole, centred over a blurred copy of itself. When the shot changes mid-clip, the layout switches at the cut.
  • Captions and titles: burned-in captions from the transcript, one line at a time, and an on-screen title for the first seconds. Both can be switched off per clip.
  • Review, then render: trim in half-second steps, retitle, drop or restore, render one clip or all of them, download.
  • Agent-ready: a REST API and an agent.md, so an AI agent can take a video from upload to finished clips without a human in the loop.

How it works

  1. The video is uploaded to the platform's media service, which transcodes it, transcribes it and gets it ready to be watched.
  2. The platform's video analysis watches and listens to the whole video in one pass, with its transcript, and proposes moments with a title, the hook and one sentence on why each works. A sound demo or a striking picture can carry a moment, not only speech. It also says how each stretch is shot: a person filling the frame, or a screen.
  3. Cut points that land mid-word move to the word's edge; cut points in silence stay put.
  4. The clip is rendered to a 1080×1920 MP4 on the managed edit service and stored in the app.

Run it locally

pnpm install
pnpm dev

pnpm dev runs clawnify dev, which generates the local Worker config in .clawnify/, creates a local database from schema.sql, and serves the app at http://localhost:5173.

Everything runs on the platform's managed services, which need an org token. Put it in .clawnify/.dev.vars, next to the generated config (that is where the local server reads it), then restart pnpm dev:

printf 'CLAWNIFY_TOKEN=clw_...\n' > .clawnify/.dev.vars

pnpm test runs the unit tests for the transcript, layout and edit logic.

License

MIT