Every feature. Zero overhead.

Stills

Screenshots, and the tools around them

A region you drag, a single window, or the whole display — captured at native backing resolution, straight to PNG. Set a countdown when the thing you want only appears after you let go of the mouse, and turn on Copy to clipboard on capture to skip the file entirely.

Scrolling capture is the one thing the App Store build cannot do: driving the page needs the Accessibility permission, which a sandboxed app may not hold, so it refuses up front and points you at `screenshot --region` instead. It handles what doesn't fit on screen: point it at a scrollable region and it auto-scrolls, grabbing and stitching the frames into one tall PNG — a whole web page, chat log, or document in a single image.

Two more live in the same row. The eyedropper samples any pixel on screen and hands you its HEX and RGB, and Pin image floats a still on top of everything so you can keep a reference in view while you work. Annotate opens the captured still in the same editor the videos use — arrows, boxes, blurs, text.

# a region, a window, or the whole display
wispframe screenshot --region 100,100,1200,800 --out shot.png
wispframe screenshot --window 6660 --out window.png
wispframe screenshot --full --delay 3 --out screen.png

# a long page, stitched into one tall PNG
wispframe scroll-capture --region 200,120,1400,900 --out page.png

# sample a pixel, straight to the clipboard
wispframe pick-color --at 640,480 --clipboard

✦ New — Subtitler

Automatic subtitles, on-device

Wispframe transcribes your video's speech with an on-device Whisper model — running on Apple Silicon's Metal, so your audio never leaves your Mac. Pick a model (large-v3-turbo by default, right down to a tiny base), and it writes word-accurate captions with a timestamp on every line.

Style them to match what the short-form apps do: 22 one-click presets — from clean captions to word-by-word popups (Pop Word), karaoke fills, soft gradient pills (Aurora, Sunset, Ocean), Neon, Serif, Script, Impact and more — then tweak every field individually: font, weight, size, gradient text and gradient background boxes, outline, shadow, per-line and per-word pills, position, and entrance animation (pop, bounce, slide, typewriter). A colored pill tracks the spoken word, the word fills karaoke-style as it's said, and you can drag the caption to move it or scroll to resize it right on the preview.

Burn the captions into the video, or export a clean SRT or VTT. Everything is scriptable from the CLI and MCP too.

# transcribe speech to an SRT (on-device Whisper)
wispframe transcribe --in talk.mp4 --out talk.srt --lang en --words

# burn styled captions in, or export SRT <-> VTT
wispframe subtitle --in talk.mp4 --captions talk.srt \
  --out talk-sub.mp4 --preset Karaoke

Audio

System audio and microphone, mixed live

Most recorders make you choose: the app's sound or your voice. Here you toggle both, set per-source gain from 0–200%, and get a single clean AAC track muxed into the file.

No virtual audio drivers, no BlackHole, no routing utilities. It's two checkboxes and two sliders.

wispframe record \
  --system-audio --mic \
  --system-audio-gain 0.5 --mic-gain 1.5 \
  --out narrated-demo.mp4

Compositing

2–9 sources on one canvas

Record live, too: pick 2–9 displays, windows, and apps in the app, hit Record, and they're captured in parallel — the moment you stop, they merge into one video in your chosen layout, carrying source ①'s audio.

Or combine existing recordings into a grid, side-by-side strip, rows, picture-in-picture corners, or a speaker layout with one flag. Perfect for multi-cam tutorials, panel recordings, and before/after comparisons.

You can also composite two videos at an exact pixel rect and time window — with the base video's audio passing through untouched.

And when a preset isn't enough, go full video mixer: the Custom layout lets you drag and resize every source anywhere on the canvas — overlap them, set per-source opacity, and higher-numbered sources draw on top. Dial the canvas up to 8K, and size the picture-in-picture main and corner tile independently.

New: the arrangement doesn't have to stay fixed. Drop layout keyframes on the timeline — side-by-side for the intro, grid at 12 seconds, one source big for the demo — and choose how each change arrives: a hard cut, a slide, an eased glide, or a cross-dissolve. Everything aspect-fits while it moves, so nothing ever distorts mid-transition.

Shape any source: render the webcam as a circle bubble, or any cell as a rounded rect, triangle, or hexagon — in the app's shape picker and headless with --shape 1:circle. And the whole composition saves as a settings file you can re-import, share, or execute in one command: wispframe run compose-settings.json.

wispframe compose-grid --sources a.mp4,b.mp4,c.mp4,d.mp4 \
  --layout grid --out quad.mp4

wispframe compose-grid --sources main.mp4,cam.mp4 --out mix.mp4 \
  --place 0:0,0,1920,1080 --place 1:1200,600,600,400 --opacity 1:60

wispframe compose --base screen.mp4 --overlay cam.mp4 \
  --rect 100,50,640,360 --start-at 12.5 --end-at 30 --out pip.mp4

# switch the arrangement mid-video, with transitions
wispframe compose-grid --sources a.mp4,b.mp4 --out show.mp4 \
  --layout-key 0:sbs --layout-key 12:grid:ease:0.6 \
  --layout-key 40:rows:fade:0.8

Reliability

Crash-proof by construction

Recordings are written as fragmented MP4, finalized every two seconds. If your Mac kernel-panics, loses power, or you kill -9 the process at minute 47, the file on disk still opens and plays — you lose at most the last two seconds.

There is no recovery step, no repair tool, no corrupted-file dialog. The file was always valid.

Export

GIFs and multi-codec output

Turn any clip into a share-ready GIF with controllable width and frame rate — from the CLI or one click in the GUI's Done row. Record in H.264 for compatibility, HEVC for size, or ProRes for editing.

wispframe to-gif --in clip.mp4 --out clip.gif --width 480 --fps 15

Workflow

Gets out of your way

Hit Record and the window disappears; a menu-bar timer is the only sign anything is happening. Stop from anywhere with ⌘⇧. — even with every window hidden. It uses less than 1% CPU, so it never gets in the way of whatever you're recording.

Run several recordings in parallel and watch them in a live session panel. Background removal (Vision on the Neural Engine, or U²-Net via ONNX) runs automatically right after recording, or on any existing video.

# record with automatic background removal
wispframe record --mask-bg vision --out out.mp4

# or clean up an existing video
wispframe remove-bg --in input.mp4 --out nobg.mp4 --backend onnx