Landing page
1/5
ShortPulse
A local-first AI agent that turns a topic into a ready-to-post vertical video: script,
voiceover, word-synced captions, visuals and music, assembled by a pipeline you can
actually read. Self-host it for free, or pay per video — no subscription either way.
- 9languages
- 6art styles
- MITlicensed
- Localor hosted
One topic in, a finished 1080×1920 MP4 out — script, voiceover, word-synced captions, visuals and ducked music, with no watermark and no subscription.
What it does
- Writes the scene breakdown, the hook and the call to action from a single line of input — or splits a script you already have into scenes.
- Generates the voiceover, then aligns the captions to it at word level, so the text lands on the syllable instead of the sentence.
- Fills each scene with stock footage or renders AI stills in one of six art styles, then composes everything in FFmpeg.
- Streams progress over a WebSocket stage by stage, so you watch the render happen instead of waiting on a spinner.
- Hands back the scene breakdown next to the player: click a scene to jump the video to it, with per-scene timings taken from the word-level transcript.
How it ships
- One codebase, two ways to run it: self-host the whole pipeline locally for nothing, or use the hosted service and pay per video through Stripe.
- MIT licensed and open source — the pipeline is meant to be read, forked and changed, not treated as a black box.
- Runs on a Hetzner box behind Caddy, with Supabase for auth and data.
