A screen recorder for modern Linux desktops (Wayland & X11).
takesy records your screen and turns it into a polished screencast: it automatically zooms in on wherever you're working, smooths the cursor motion, and composites the result — padded background, rounded corners, drop shadow — into a finished mp4. You just record; takesy directs the shot.
- Tracked camera — a smooth, spring-damped pan and zoom that follows where the screen is changing, breathing tighter or wider as activity concentrates.
- Never crops the action — watches the screen (frame diffs) and fits the zoom to the active region, so relevant changes always stay in frame.
- Smoothed cursor — a speed-adaptive spring that aims at your clicks (cutting the usual overshoot) yet stays responsive when you point at things.
- Polished compositing — the screen inset on a padded background with rounded corners (optional) and a soft drop shadow.
- Custom cursor — draw your own cursor image instead of the built-in arrow.
- Click ripples — expanding rings on click, with a subtle cursor press animation.
- Trim idle time — cut dead stretches (no screen change) to tighten long demos.
- Webcam inset — record your camera live during capture (
--webcam; the camera is chosen in the preview), or composite a pre-recorded clip, as a picture-in-picture — its corner radius sweeps from circle through rounded to square, with configurable border width and colour. - Webcam framing preview — whenever you record a live webcam, takesy first opens a local web page to pick the input camera and frame yourself (drag to pan, scroll to zoom); what you see is what gets recorded.
- Countdown — a 3-2-1 (with audible beeps) before recording starts, so you can get ready.
- Pause / resume —
kill -USR1 <pid>toggles capture; paused time is cut, timeline stays gapless. - Audio — optionally record desktop sound, your mic, or both, muxed into the mp4.
- Capture once, render many — record once, then re-render with different framing, background, or cursor without re-recording.
- Click to stop — record for as long as you like; finish with your desktop's screencast Stop button. Frames stream to a compressed intermediate during capture and straight to the encoder on render, so long recordings stay small on disk.
- One self-contained binary — builds to a single
takesyexecutable.
- A modern Linux desktop (Wayland or X11) with
xdg-desktop-portal+ PipeWire (GNOME, KDE, or wlroots) ffmpegwith an H.264 encoder (libx264orlibopenh264)pactl(PulseAudio/PipeWire) — only for--audio- SBCL and ocicl to build (dependencies resolve automatically)
Prebuilt x86_64 packages are published for each release. The packages are
GPG-signed and pull in ffmpeg and the PipeWire/EGL runtime libraries takesy
needs.
Add the takesy repository and install:
sudo dnf config-manager addrepo --from-repofile=https://atgreen.github.io/takesy/rpm-repo/takesy.repo
sudo dnf install takesyThe signing key is imported automatically by dnf on first install.
Add the takesy repository and install:
curl -fsSL https://atgreen.github.io/takesy/deb-repo/takesy-archive-keyring.gpg | sudo tee /usr/share/keyrings/takesy-archive-keyring.asc > /dev/null
echo "deb [signed-by=/usr/share/keyrings/takesy-archive-keyring.asc] https://atgreen.github.io/takesy/deb-repo stable main" | sudo tee /etc/apt/sources.list.d/takesy.list
sudo apt update
sudo apt install takesyIndividual .rpm / .deb files are also attached to each
GitHub release.
make # -> ./takesy
make install # copies takesy to ~/.local/bintakesy # record; click Stop in the top bar to finish
takesy --output talk.mp4 --duration 20
takesy --audio both # record desktop sound + mic too
takesy --webcam # picture-in-picture from your webcam (pick it in the preview)
takesy --cursor arrow.png # use a custom cursor image
takesy --bg-image wallpaper.png # put an image behind the inset
takesy --output demo.gif # straight to an animated GIF (no audio)
takesy helpPick a source in the screen-share dialog, do your thing, then click your
desktop's screencast Stop button to finish. The cursor is hidden during
capture — takesy needs its position to drive the auto-zoom — and restored when
you're done. (If you haven't run make install, invoke it as ./takesy.)
The pipeline splits so you can record once and re-render with different framing as often as you like — no re-recording:
takesy capture --dir /tmp/talk # capture only -> self-contained recording dir
takesy render /tmp/talk --output a.mp4 --bg dark
takesy render /tmp/talk --output b.mp4 --bg navy --aspect 9:16 # same capture, new lookcapture writes the frames plus a manifest into the dir; render runs the
auto-zoom and compositing over it, honouring all the options below.
| Option | Meaning | Default |
|---|---|---|
--output PATH |
output file; a .gif path produces an animated GIF (no audio) instead of an mp4 |
/tmp/takesy-record.mp4 |
--duration S |
optional maximum length (safety cap); by default there's none — you set the length by clicking Stop | none |
--fps N |
output frame rate | 24 |
--height N |
maximum output height in px (never upscales past the content) | 1200 |
--bg COLOR |
background — blur (default; a frosted copy of the screen), or #RRGGBB / a name (black, white, dark, navy, slate, …) for a solid |
blur |
--bg-image PATH |
background image (PNG or anything ffmpeg reads), cover-fit behind the inset; overrides --bg |
— |
--aspect W:H |
reframe the output to an aspect (e.g. 9:16 vertical, 1:1 square); the screen is scaled to fit inside it (letterboxed, never cropped) |
content aspect |
--region X,Y,W,H |
frame a fixed screen region (source pixels) instead of auto-cropping to content | auto-crop |
--trim-idle |
cut idle stretches (no screen change) longer than --idle-threshold down to --max-idle seconds; tightens long demos (drops audio for now) |
off |
--ripples on|off |
expanding ring on each click, with a subtle cursor press animation | on |
--webcam |
record a live webcam inset — the camera and framing are chosen in the preview that opens before recording (no device to name). For takesy render, --webcam FILE instead composites a pre-recorded video/image |
off |
--webcam-corner F |
inset corner radius, fraction of half-size: 1 = circle, 0 = square, between = rounded |
1 |
--webcam-border F |
inset border width, fraction of the inset (0 = none) |
0.012 |
--webcam-border-color C |
inset border colour — a name or #RRGGBB |
white |
--webcam-pos, --webcam-size |
inset corner (br|bl|tr|tl) and diameter (fraction of height) |
br, 0.22 |
--webcam-zoom, --webcam-pan-x, --webcam-pan-y |
webcam framing — normally set in the auto preview, but can be forced | 1, 0, 0 |
--corner-radius F |
rounded-corner radius (fraction of the content's shorter side); 0 gives square corners |
0.09 |
--margin F |
inset margin around the screen, as a fraction of the frame | 0.04 |
--cursor PATH |
draw a custom cursor image (PNG or anything ffmpeg reads) instead of the built-in arrow | built-in arrow |
--cursor-hotspot X,Y |
the image's click point, as a fraction of its size | 0,0 (top-left) |
--cursor-size F |
cursor height as a fraction of the output height | 0.06 |
--audio MODE |
record audio: system (desktop/monitor), mic, or both (mixed). Muxed into the mp4 as AAC; set at capture/record time |
off |
--audio-offset S |
A/V sync nudge (seconds, signed); + pushes audio later to fix audio that leads the picture. Auto-aligned by the count-in clapperboard, so normally 0 |
0 |
--webcam-offset S |
webcam-PiP sync nudge (seconds, signed); + pushes the webcam later. Auto-aligned from the capture clock, so normally 0 |
0 |
--countdown N |
seconds of 3-2-1 countdown before recording starts; set at capture/record time |
3 |
takesy directs the camera like an editor rather than following the cursor: it
ranks evidence (clicks > sustained activity > cursor), picks shots from a small
vocabulary (overview / working / detail), holds each shot, and moves in single
composed, eased gestures — establishing wide first, widening through the overview
on large jumps, and never darting after a brief flash. There are no per-move
knobs to dial; the direction is automatic. --log-camera PATH writes the
per-frame camera path (zoom/pan) to CSV and prints a motion summary if you want
to inspect it.
Cursor feel and how much the camera reacts to the pointer are tunable:
| Option | Meaning | Default |
|---|---|---|
--cursor-omega-fast R |
cursor-spring stiffness when the pointer moves fast — higher tracks the real pointer more tightly (less lag) | 60.0 |
--cursor-anticipate S |
seconds before a click to start aiming the cursor straight at it | 0.15 |
--camera-smoothing S |
smooth hand tremor out of the auto-zoom camera, seconds (0 = off) |
0.15 |
--damage-anchor MODE |
how the camera frames screen activity: reading keeps the left column / top context in view; center centres on the change |
reading |
--speech-aware |
when narration is recorded, land camera moves on speech pauses (cut on the breath, not mid-sentence) | on |
- Capture (
takesy capture) — receives frames and the cursor position overxdg-desktop-portal+ PipeWire, streaming them to a compressed intermediate, plus optional audio. Writes a self-contained recording dir. - Direct — detects where you focused, plans eased auto-zoom keyframes fit to the active region, and smooths the cursor path.
- Composite (
takesy render, GPU) — renders the zoom, background, rounded corners, drop shadow, and cursor overlay with EGL/OpenGL, streaming frames straight toffmpeg(and muxing audio) — no large scratch files.
takesy (or takesy record) runs all three in one shot; capture and render
split them so one recording can be rendered many ways.
Created by Anthony Green; distributed under the GNU General Public License v3.0 or later.