Shorts

The CapCut-style editor for short-form video. Free, open source, and it runs entirely on your machine.

Auto captions · edit by deleting words · one-click silence and retake removal · export ready for Reels, TikTok and Shorts

Install · Features · Why · FAQ

_{Real recording, with the transcription and export waits sped up. MP4 version. The clip is synthetic, so no personal footage ships with this repo.}

From raw recording to posted in about two minutes

1. Drop in your video. Phone footage, screen recordings, anything ffmpeg reads.
2. Click Auto captions. Phonon-2 transcribes only what's on your timeline, on your own CPU.
3. Click Cut retakes and Remove silences. False starts and dead air are gone.
4. Pick a caption style. Word-by-word highlights, boxes, karaoke, one word at a time. Drag it where you want.
5. Delete words in the transcript to cut them from the video, like editing a doc.
6. Export. 1080×1920 H.264 at −14 LUFS, plus .srt captions and a .txt transcript for your post.

Why I built it

I make short-form videos for IdealJob, and Shorts is the editor I built to edit them. Before, I used CapCut or DaVinci Resolve. On Linux neither works well:

• CapCut has no Linux version.
• DaVinci Resolve runs on Linux, but the experience is miserable for short-form content:
  • The free version can't read H.264 or AAC on Linux, so every phone or screen recording had to be converted to DNxHR before I could import it.
  • Captions meant exporting the audio and running a separate transcription tool in the terminal.
  • Every export needed another pass through ffmpeg before Instagram or TikTok would accept it.
  • Caption styling was very limited.

So I built my own: the editing flow I liked in CapCut, running natively on Linux (and now macOS and Windows), with everything, transcription included, on my own machine.

How it compares

	Shorts	CapCut	DaVinci Resolve (free)
Runs on Linux	✅	❌	✅
Opens phone footage (H.264/AAC) on Linux	✅	—	❌
Auto captions	✅ on your CPU	✅	❌ Studio only
Edit the video by editing the transcript	✅	✅	❌ Studio only
Open source	✅ MIT	❌	❌
Price	Free	Free + Pro subscription	Free / Studio $295

Features

Speech editing

• Auto captions transcribe only the parts of your videos that are on the timeline, and only the new parts after you edit.
• Transcript editing: delete words to cut them, fix misheard words, highlight words for captions.
• One-click Remove silences, Cut retakes (false starts), Cut fillers (um, uh) and Punch-in (alternating zoom on cuts).

Captions and graphics

• 7 caption styles (word highlight, box, one word, karaoke…) with full control over font, colors, outline, shadow and animation. Drag them on the canvas.
• Text presets, emoji stickers, and highlight shapes (boxes, circles, arrows, underlines).

Timeline

• A main track plus as many video, image, text and audio rows as you need. Free placement or a magnetic main track.
• Trim, split, delete left/right of the playhead, duplicate, copy/paste, markers, snapping. Zoom stays anchored on the playhead.
• Transitions (dissolve, dips, slides, wipes, zoom, circle) and keyframes for position, scale, rotation and opacity.
• Speed, flip, freeze frame, detach audio, slow zoom (Ken Burns), intro effects.
• Smooth cuts: hides jump cuts by lining your face up across the cut, then gliding back to the normal framing.

Look and sound

• Color looks and adjustments: brightness, contrast, saturation, temperature, sharpen, denoise.
• Clean surface: draw a box over a desk or mat and dust, lint and crumbs disappear.
• Phone mockups: put any video inside a phone (black or silver) as an element you can move, resize, rotate and animate, or frame the whole video in a phone on a gradient, colour or blurred background.
• Auto reframe follows your face to turn landscape footage into 9:16. Background replacement with blur or a color.
• Music and sound effects that duck under your voice automatically, plus voiceover recording.
• Loudness normalization to −14 LUFS, voice enhance, automatic repair of clipped (distorted) audio, and Lower breaths, which makes breathing between sentences almost silent without changing your timing.

Export

• 30 or 60 fps, 1080p or 720p, H.264 + AAC. Uses the Intel GPU encoder (VA-API) on Linux when available, x264 otherwise.
• No generational loss: each clip is rendered losslessly once and cached; only the final encode compresses.

Install

You need git. The setup script tells you exactly what else to install (FFmpeg, uv and Bun) if anything is missing.

Linux and macOS

git clone https://github.com/MarioRodrigues10/shorts.git
cd shorts
scripts/setup.sh

On macOS, get the requirements with Homebrew: brew install ffmpeg uv oven-sh/bun/bun.

Windows 10 / 11 (PowerShell)

git clone https://github.com/MarioRodrigues10/shorts.git
cd shorts
powershell -ExecutionPolicy Bypass -File scripts\setup.ps1

On Windows, missing tools are installed for you with winget.

The setup script:

• creates the Python environments
• downloads the caption fonts and vision models (all openly licensed)
• builds the UI
• installs the shorts command, plus a background service on Linux, ~/Applications/Shorts.command on macOS, or a Start Menu shortcut on Windows

The Phonon-2 speech model (about 170 MB) downloads on your first transcription. Add --no-vision (-NoVision on Windows) to skip auto reframe and background replacement and save about 400 MB.

Use

shorts                    # open the app in your browser
shorts path/to/clip.mp4   # start a project from a clip
shorts convert file.mov   # platform-ready MP4, no editing

Press ? in the editor for every shortcut. The essentials:

Key	Action	Key	Action
Space	Play / pause	S	Split
Q / W	Delete left / right of playhead	Del	Delete
← →	Frame step (Shift: 1 s)	↑ ↓	Previous / next cut
T	Add text	Ctrl+Z / Y	Undo / redo
Ctrl+E	Export	?	All shortcuts

Coming from DaVinci Resolve? Copy integrations/resolve/Send to Shorts.py into ~/.local/share/DaVinciResolve/Fusion/Scripts/Utility/. Then Workspace → Scripts → Send to Shorts renders the current timeline and opens it here for captions and export.

Meet Shorty

Shorty keeps you company in the app: listening while it transcribes, celebrating when your video is ready, and napping when there's nothing to edit yet. It's a clapperboard with scissors, because editing is mostly cutting. Every pose is plain SVG generated by docs/mascot/build.py.

Your data stays yours

Transcription, rendering and the editor all run on your computer. No accounts, no uploads, no API keys, no telemetry.

What	Where
Settings	~/.config/shorts-studio/config.json
Projects, transcripts, proxies, render cache	~/.local/share/shorts-studio/
Imported files	~/Videos/Shorts/Media (configurable)
Exports	~/Videos/Shorts/Exports (configurable)

config.json lets you change:

• the folders shown in the media browser (library_dirs)
• upload_dir and export_dir
• the transcription model and language
• vocabulary (comma-separated names and terms the model should know)
• encoder settings

FAQ

Does it work in my language? Yes. Set your "language" (for example "pt" or "es") in config.json. English uses Phonon-2 (the default "model"), which is fast and accurate. Other languages use Whisper's multilingual small model, or any Whisper model you set as "model".

Do I need a GPU? No. Everything runs on the CPU, and it was built on an 8 GB laptop. Previews use lightweight proxies, heavy jobs run one at a time at low priority, and on Linux the service is memory-capped. Intel GPUs speed up exports on Linux.

Is it really offline? After setup, yes. The only downloads are the speech model on first use and the models and fonts during setup.

Which platforms are tested? Linux is the daily driver. CI runs the setup scripts and an export on Linux, macOS and Windows on every push. Bug reports from macOS and Windows are very welcome.

How it works

web/            React + Tailwind + Radix UI, built into static/app
  src/core/     the editing engine: project model, timeline, preview player,
                caption/text renderer, keyframes (plain JS)
server.py       stdlib HTTP server on 127.0.0.1: media registry, background worker
                (proxies, filmstrips, waveforms, transcripts), projects, export via ffmpeg
transcribe.py   Phonon-2 (or faster-whisper) on CPU + energy-based word boundary refinement
vision.py       MediaPipe face tracking and person segmentation (CPU)

• Preview and export match. The same JavaScript draws captions and text in the preview and in the export. The browser renders the caption layer and the server composites it.
• No drift across hundreds of cuts. Each clip is rendered to a lossless cached piece on a single frame grid. Transitions are separate pieces. The final pass adds overlays, the caption layer and the audio mix (ducking, two-pass loudness), then encodes once.

Contributing

Issues and pull requests are welcome. See CONTRIBUTING.md for the dev setup, how to test without touching your own projects, and how the repo keeps personal data out of git. scripts/demo/record.sh re-records the demo above from a synthetic clip.

If Shorts saves you from a Linux editing headache, a ⭐ helps other people find it.

License

MIT © Mário Rodrigues. Made to edit the videos for IdealJob.

Built with Phonon-2 by Fermion Research (CC BY 4.0), faster-whisper, MediaPipe, FFmpeg, React, Tailwind CSS, Radix UI and Phosphor Icons. Caption fonts from Google Fonts (SIL Open Font License).