Self-hosted video repurposing

Long videos in.
Clips worth posting out.

Clip finds the moments that hold attention, reframes them for vertical screens and captions them word by word — on your server, with your storage and the AI model you choose.

3
output formats
7
scoring dimensions
0
subscriptions
founders-podcast-ep42.mp4
6 moments
64
91
58
77
83
52
0:0018:0036:0054:001:12:48

15:07 → 15:49 · 42s

“We almost shut the company down — then one customer email changed everything.”

91

score

Hook

Context

Emotion

Payoff

one customer email changed
9:16face-tracked

From a two-hour recording to a week of posts.

  1. 01

    Upload

    Drop in an episode, a talk or a stream. Files go straight to your private R2 bucket — resumable, up to 20 GB.

  2. 02

    Transcribe

    Word-level timestamps with faster-whisper, on your own server. No audio leaves your infrastructure.

  3. 03

    Score

    Every window is ranked on hook, context, emotion, information, story, pacing and payoff — weights you control.

  4. 04

    Edit

    Trim on word boundaries, adjust the crop, restyle captions. Every change is a version you can go back to.

  5. 05

    Ship

    Render vertical, square or landscape and download — or publish to the accounts you've connected.

Clip is in active development. Projects, uploads, teams and AI provider settings work today; features marked “In progress” are being brought over from the OpenShorts pipeline milestone by milestone.

In progress

Moments that stand on their own

Clips start on a hook and end on a payoff, never mid-sentence. Model timestamps are checked against the real audio before you see them.

In progress

Reframing that follows the speaker

Face and speaker tracking with a steady, tripod-like camera. Split screen for two people, screen-plus-presenter for demos, manual override anytime.

In progress

Captions people actually read

Word-by-word highlighting, reusable templates, outlines, boxes and position control. Export SRT and VTT too.

In progress

An editor, not a black box

A timeline built on a versioned project format. Undo, autosave and render previews before the final export.

In progress

B-roll with the paperwork done

Suggestions from the transcript, matched against licensed stock. Source, licence and attribution stay attached to every asset.

In progress

Publish where you're approved

YouTube, Instagram, TikTok and Facebook through their official APIs — with a download button for everything else.

Your footage stays yours.

Clip runs on your own server. There are no plans, credits or usage caps — only the limits you set to keep the machine healthy.

  • Private storage

    Originals and renders live in a private Cloudflare R2 bucket, shared only through short-lived signed links.

  • Any model, or none

    Claude, OpenAI, Gemini, any OpenAI-compatible server or a local Ollama. Basic processing works without an AI key.

  • Teams and roles

    Organizations with owner, admin, editor and viewer roles, enforced by the API and by the database itself.

  • Built on open source

    Clip extends OpenShorts (MIT) with its own platform, editor and storage layer.