Glyph Cast

Screen videos, cleaned up. Record your screen on a Mac and get back a clean MP4: the voice cleaned and leveled, the ums and dead air cut, the screen framed on a background with zooms where you click, an intro and outro attached if you want them. Screen recording and framing plus transcript editing, in one free Mac app.

Its page is glyph.fru.dev/cast. Part of the Glyph family by fru.dev: Glyph Studio, Glyph Jot, Glyph Ibis and Glyph Rig, plus two tools, Glyph Arena and Glyph Cast. Formerly called Capture.

Install

1. Download Glyph-Cast.dmg and drag Glyph Cast to Applications. It is signed and notarized by Apple.
2. Install the two tools it runs, once: brew install ffmpeg whisper-cpp
3. Open it. macOS asks once for Screen Recording and the Microphone. If Glyph Cast is missing from System Settings > Privacy & Security > Screen & System Audio Recording, click + there, add Glyph Cast, switch it on and click Reopen in the app (it shows a banner with both steps).

The first recording downloads the models (about 180 MB) to ~/Library/Application Support/Capture/models: whisper base.en, the Silero voice detector, DeepFilterNet's deep-filter and an RNNoise model.

Needs macOS 15 or later on Apple silicon.

Use

1. Pick the display and microphone (the meter beside it moves when you speak) and press Start recording, or Option-Command-R from anywhere. A 3-2-1 countdown runs and the window gets out of the way. While you record, a bar at the top of the screen shows REC, the time, your mic live, a mic picker (switch mid-recording), Pause and Resume (Option-Command-P), Restart (the take goes to the Trash and the countdown runs again) and Stop. A thin red edge runs round the recorded screen. Neither the bar nor the edge is in the video, and paused time is left out.

2. Press Option-Command-R again (or Stop). Glyph Cast prepares the recording on its own, about a minute for every ten recorded:

   • lines the mic up with the video (same host clock, exact),
   • removes background noise (DeepFilterNet; RNNoise if it can't run),
   • shapes the voice (rumble cut, presence lift, de-ess, gentle compression) and levels it to -14 LUFS, YouTube's loudness, in two passes,
   • transcribes it with whisper, tightening each word's timing against the real silences,
   • finds filler words (um, uh, hmm; repeated words like "the the"; like, you know, I mean) and long pauses. Nothing is cut yet: fillers show highlighted in the transcript.

3. Clean up with the buttons above the transcript. Each is one click, and a second click undoes it:

   • Clean up everything: all three below at once,
   • Remove filler words (with a count; the arrow picks which kinds),
   • Shorten pauses (the arrow sets how long a pause may be, and keeps pauses while the screen was moving, so silent demo steps stay in),
   • Studio Sound: noise removed, voice shaped, leveled to -14 LUFS; the arrow's Intensity blends your original mic back in.

   Cuts follow the voice's loudness, not only whisper's word times (which run about 0.1 s late): a removed filler goes from its first sound to its last, leaving a natural 0.25 s pause. The waveform under the player shows what is cut in red; click it to play from there. Right-click a word to cut or keep it, or every "uh" at once. "Clean up new recordings automatically" in the inspector does all of it as soon as a recording is ready.

4. Fine-tune in the transcript if you want: click a word to play from there; Option-click a word to cut it or bring it back; click one word, Shift-click another, then Cut (or Delete) or Restore; click a pause chip (2.4s) to cut or keep that pause; Reset edits goes back to the automatic cut.

5. The inspector on the right is remembered for every recording after: export size (1080p, 1440p, 4K), the background (18 wave wallpapers drawn for the app, your desktop picture, any image, or no frame), padding, corners, shadow, zoom on clicks and how far, click rings, and intro and outro clips. The preview updates as you change them.

6. Export MP4. The file goes to ~/Movies/Capture/Exports/<recording>.mp4 with a matching .srt of the kept words: H.264 High, 60 fps, 12 Mbps at 1080p (24 at 1440p, 45 at 4K), AAC 192 kbps stereo.

Recordings live in ~/Movies/Capture/<date time>/: screen.mov (HEVC, full Retina resolution), mic.caf, system.caf (if you turned on system sound), events.json (cursor, clicks, screen activity), edit.json (your cuts and cleanup switches) and work/ (cleaned voice, original voice for the Studio Sound blend, transcript). Nothing leaves your Mac.

The wallpapers are drawn by scripts/make_wallpapers.py (layered silk waves in gradients) into assets/wallpapers/. Edit a palette and run it again to change them.

Build

brew install ffmpeg whisper-cpp
./scripts/build.sh --install --run     # builds "Glyph Cast.app", signs it, copies it to /Applications, opens it

The bundle id is dev.fru.capture and the data folders keep the Capture name, so an install of the older Capture keeps its permissions, settings and recordings.

Test

swift build
.build/debug/Capture --selftest              # a fake recording with ums and pauses, through process + export
.build/debug/Capture --record-test 8 --speak # records the real screen and mic for 8 s, speaks a line
CAPTURE_ROOT=<folder of recordings> .build/debug/Capture --open "<recording>" --uitest <out dir> [--record-flow] [--appearance dark]
.build/release/Capture --process "<recording>" [--export-only] [--height 1080] [--out <folder>]

--selftest checks the transcript, filler kinds (um, repeats, tics), that nothing is cut until you clean up, that fillers and a dead pause are then cut and a pause with a click is kept, -14 LUFS, the export size and length (intro + kept + outro), the frame on its background, captions, the Studio Sound blend and that system sound reaches the edit, and then transcribes the export to confirm no filler is left and no real word was clipped.

--uitest drives the real window: it finds each control through the app's own accessibility tree and presses it. It covers the cleanup buttons, Option-click, Shift-select + Cut, Reset, pause chips, the waveform, wallpapers and a pixel check of the redrawn preview, toggles, the size picker, Export, the record page and the shortcut. --record-flow adds the whole loop: Record button, countdown, recording, stop, processing, editor. Screenshots of each step go to the out dir.

How it is built

Swift, no packages. ScreenCaptureKit for screen, mic and system sound; AVFoundation to write and to compose; Core Image for every frame; ffmpeg, deep-filter and whisper-cli as tools.

File	What it does
Recorder.swift	SCStream to HEVC screen.mov + float mic.caf/system.caf; cursor, clicks, changed-screen area
Processor.swift	align, denoise, voice chain, loudness, transcribe, find fillers and pauses, edit.json
Project.swift	the files, the edit model, the cut rules, Style (settings)
Composer.swift	cuts into an AVComposition, the per-frame renderer (background, shadow, corners, zoom, rings), export, SRT
App.swift, Views.swift, HUD.swift	window, menu bar, record page, editor, transcript, inspector, recording bar
SelfTest.swift, UITest.swift	the tests above

Not built yet: camera bubble, captions burned into the video, a re-rendered smooth cursor, a music bed.

Versions

• 0.1.2 (2026-10-25): name a recording: double-click its name in the list (or right-click > Rename), or press the pencil on the recording bar while it records. The name names the exported MP4. "Add intro and outro" in the Export section leaves them off when they get added later. Glyph Cast now shows up in System Settings > Screen & System Audio Recording on macOS 15 and later.
• 0.1.1 (2026-10-25): Capture is now Glyph Cast, in the Glyph family: new name and icon, a signed and notarized download. Same bundle id and folders, so settings and recordings carry over.

License

MIT. See LICENSE.