Gerty

A browser-based, open source, free to use video editor.

• No login, no personalised tracking, no ads, no scraping your data, no AI, no bullshit.
• Code is fully open source
• Import images, video clips, audio clips
• Add custom arrows, shapes, text, all with animated effects, keyframing
• Add video-wide effects, like a camera zoom, sepia, vignette, film grain, etc.
• Preview the video in real time. Export to MP4, with customisable quality/size

"What it lacks in features, it makes up for in not being dogshit" - me

https://gerty.tomg.cool/ 👈 use it right now

VideoEditor-Demo.mp4

---

What you can do

• Bring in your media - drop in images, video clips and audio. Everything stays on your machine; nothing is uploaded.
• Build on a timeline - stack clips on layers, move and group them, trim their ends, and split a clip in two at the playhead.
• Annotate - arrows, text boxes, rectangles, circles and freehand pen, all fully stylable.
• Animate anything - keyframe an object's position, size, rotation and opacity; add on-appear / on-exit transitions; give text a type-on reveal and arrows a draw-on.
• Add a camera - screen-recorder-style zoom and pan push-ins that glide across the frame, with their own keyframes.
• Grade and stylise - a deep effects menu: black & white, sepia, vignette, film grain, old film, hue shift, contrast, bleach bypass, light leaks, chromatic split, plus GPU shader looks like gradient maps, duotone, CRT, VHS, halftone and comic ink.
• One-click presets - drop a whole graded look on a clip: Cinematic (teal & orange), Cinematic Cool, Super 8, Retro TV, Film Noir, Comic Book and more. Every preset is just a starting point you can tweak.
• Mark the beat - tap M while it plays to drop markers, then snap edits to them.
• Preview live, then export - real-time preview, then export to MP4 (H.264 + AAC) at a resolution and quality you choose, with a file-size estimate before you commit.
• Save your work - export a project to a .gerty file and re-import it later to pick up where you left off.

Keyboard shortcuts

Key	Action
Space	Play / pause
V	Toggle camera Live / Frame view
M	Drop a marker at the playhead (works while playing)
, / .	Jump to previous / next marker
S	Split the selected audio/video clip at the playhead
H	Hide / show the selected object or zoom
Delete / Backspace	Delete the selection
Ctrl+Z	Undo
Ctrl+Y / Ctrl+Shift+Z	Redo
Enter / Escape	Finish the current drawing / deselect

Privacy

There is no backend. Your media never leaves the browser - imported files live in your browser's own storage (IndexedDB) and rendering, editing and export all happen locally on your device.

---

Running it locally

npm install
npm run dev

Then open the printed local URL. To make a production build:

npm run build    # type-check + bundle to dist/
npm run preview  # serve the production build locally

Tech stack

• React 19 + TypeScript, bundled with Vite
• Tailwind CSS v4 for styling
• Canvas 2D for the shared preview/export compositor, with a WebGL (regl) post-process pass for per-pixel shader effects
• WebCodecs + mp4-muxer for in-browser MP4 export (with a MediaRecorder → WebM fallback on browsers without WebCodecs)

A single pure renderFrame compositor drives both the live preview and export, so what you see is what you get. For a full architecture tour, see CLAUDE.md.

---

Known limitations & long-term to-dos

A candid running list of rough edges and deferred work. None of it blocks day-to-day use - these are the known trade-offs, kept here so they don't get lost.

Performance

• Export runs on the main thread, so the UI freezes for the duration of an export and it can't be cancelled. Moving it to a worker is the single biggest open item (spec 09).
• During playback the whole UI re-renders every frame (globalTime is React state). Fine for most projects, but it can get janky on video-heavy ones. The canvas already renders on its own rAF loop; throttling the Timeline / Properties panels is the remaining step.
• No devicePixelRatio handling - the canvas backing store is the raw project size and CSS scales it, so the preview can look a little soft on HiDPI displays.

Audio

• Preview preserves pitch when a clip's speed changes, but export does not yet (pitch shifts with the rate). See spec 11.
• General audio polish - fades, finer volume control - is still open (spec 15).

Auto-captions & text-to-speech (in-browser ML)

• Watch Cloudflare R2 Class B (read) operations. The Whisper caption model and the pocket-tts voices are self-hosted in R2 (gerty-models.tomg.cool). Egress is free, but every model-file fetch is a Class B op, so a wave of first-time caption/TTS users drives that number up. Weights are browser-cached after first load, so repeat use is cheap.
• Both the caption and TTS workers run single-threaded (numThreads = 1) for first-run reliability; revisit multi-threading for speed once it's proven stable across browsers.
• Caption accuracy rides on distil-small.en. Swapping MODEL_ID in captions.worker.ts for a larger Whisper is a one-line change if transcripts are too rough (bigger download, better accuracy).
• Caption cues are edited from the Properties panel; dragging cue bars directly on the timeline is deferred.

Editor UX

• Resize handles for objects dragged outside the canvas bounds aren't drawn - this needs a canvas→HTML overlay refactor (spec 05).
• Objects can only be edited in the camera's Frame view, not while zoomed in Live view; adding zoomed editing would require inverting the camera transform for hit-testing (spec 13).

License

Licensed under the GNU General Public License v3.0. You are free to use, study, share and modify this software; if you distribute a modified version, it must also be free software under the same license.