Free · Open source · macOS

Travel reels, cut from raw footage on your Mac.

Jauntcut turns phone and GoPro clips into finished Instagram reels. Vision models read every frame on your laptop, each edit is a small typed file, and quality gates stop a bad cut before it renders.

  • Footage never leaves your Mac
  • No account, no cloud API
  • MIT licensed

Why Jauntcut

Hand-cut reels, without the evening of editing.

Phone editing apps make every place look like the same template. Jauntcut gives each place its own type and transitions, and does the measurable work for you: timing to the beat, placing text, catching repeats and colour jumps.

Sees the footage

Subject cutouts, depth, saliency, faces, on-screen text and stabilisation, all with Apple Vision and Core ML on your Mac.

Edits are code

A reel is one small TypeScript file: clips, beats and effects. Easy to diff, version and repeat, and validated before anything renders.

Gates before render

Text over a face, the same scene twice, a dead hold or a clipped highlight fails the check, with the exact clip and the fix.

Cut to the beat

Beat and tempo analysis puts every cut on the music. Works with your own track or a guide copy of a trending Instagram sound.

11 title styles

From handwritten to film-title lockups and race lettering set into the tarmac, each built once and reused on any reel.

15 transitions

Doorway portals, match cuts, whip pans, light-order dissolves and a sketch-to-watercolour wash, plus the plain cut.

Made with Jauntcut

Every reel gets its own look.

Four reels cut from raw travel footage. Tap one to play it.

How it works

Eight stages, in a fixed order.

Each stage writes files the next one reads, so any stage can be rerun on its own, and a failed gate stops the render.

  1. Sort & prepareSort by GPS and time; normalise phone, GoPro and photo media
  2. AnalyseSubjects, depth, saliency, faces, beats
  3. EditOne typed reel.ts per reel
  4. LayoutText goes where it covers the least
  5. CheckSchema, media, timing and pacing rules
  6. QA beforeRepeats, faces under text, colour jumps
  7. RenderReact compositions to 1080×1920
  8. QA afterFrozen frames, loudness, the Instagram file spec

On-device vision

What Jauntcut sees in one frame.

Every model runs from a small Swift tool on your Mac and writes JSON or PNG for the next stage. No frame is sent anywhere.

Raw GoPro frame: a paramotor pilot in a red suit above a green valley
The raw frame. Wide GoPro footage, before any edit.
The pilot cut out from the background with an outline
Subject cutout (Apple Vision). Titles sit behind people; subjects move in front of text.
Depth map: the pilot and frame bright and warm, the valley dark
Depth (Depth Anything V2, Core ML). 3D camera moves on stills and rack focus.
Attention heat map centred on the pilot, with the detected face boxed
Saliency and faces (Vision). Text goes where the eye is not, and never on a face.

Edits as code

A reel is one small, typed file.

It lists the clips, their length in beats and every effect. A schema validates it before anything renders, so a typo or an impossible value fails in a second, not after a full render.

  • RepeatableChange one line and render again. Every version is a diff.
  • Reasons next to choicesComments record why each clip is there, so the next edit starts from that history.
  • Preview livenpm run studio opens the reel in the browser while you edit.
export const reel: ReelInput = {
  slug: "chiang-mai-circuit",
  style: "energetic",
  hook: "Not your usual Chiang Mai day",
  titleStyle: "race",
  clips: [
    // line-up POV, the hook
    { src: "IMG_2595.MOV", beats: 10, startAt: 0.2 },
    // a photo: we lift off the garage
    { src: "FF2905C1.jpg", beats: 3, motion: "push-in",
      parallax: { amount: 0.07, x: 0.5, y: 1.0 } },
    // the drop, at half speed
    { src: "IMG_2618.mov", beats: 6, startAt: 1.6,
      ramp: { atBeat: 1, speed: 0.5 } },
  ],
};

Quality gates

Nothing renders until it passes.

Each gate reports what the code actually produced. Thresholds come from measuring real footage, not defaults: the same scene shown twice measured a feature-print distance of 0.21 to 0.42 and different shots 0.49 or more, so the repeat gate fails below 0.45.

GateWhenWhat it catches
SchemaBefore anythingInvalid or impossible edit settings
checkBefore renderMissing media, upscaling past 1:1, clips running past their source, dead holds, text faster than 17 characters a second
qa --preBefore renderThe same scene twice, text over faces, colour jumps at cuts, clipped highlights, skin tones off the skin line
qa --postAfter renderBlack or frozen frames, loudness outside −16 to −10 LUFS, peaks above −1 dBFS, the Instagram file spec, leftover metadata
timing-testEngine changesRenders frames that encode their own number and reads every one back; fails on any error over one frame
regressEngine changesHashes one frame per second of every finished reel, so a new feature cannot quietly change an old reel

Roadmap

Where Jauntcut is going.

Today Jauntcut runs from Terminal and you write the edit file. The next releases make it work for anyone.

Available now

One-command install

Sets up everything on a Mac, downloads the depth model and checks the toolchain.

Next

A Mac app

Drop in your footage, pick a style, render. No Terminal and no code.

Later

Directed by a sentence

"Moody old village, slow" turned into a schema-checked edit, plus a look suggested from the music and footage.

Make your first reel today.

One command installs Jauntcut and everything it needs. Paste it into Terminal.

curl -fsSL https://jauntcut.com/install.sh | bash

macOS 14 (Sonoma) or newer. Apple silicon recommended. Read the full guide