▶ Play the live demo (browser edition — allow camera access). The browser edition now has full feature parity with the desktop app: solo & 2-player duel, best-of-3, countdown, Web-Audio sound + ambient pad, attract-mode demo, and session stats. The desktop app (below) is handy when you want a dedicated window with no browser chrome.
A webcam neck-mobility game for a culture × humanities × technology conference stall — a FACTS-H Lab installation.
Tilt your head left and right (gentle lateral neck flexion — a real mobility exercise) to glide a glowing lantern and catch the falling icons of human culture — theatre, music, manuscripts, science, art, craft. Build combos, dodge the glitch-voids, beat the stall high score. Each round is 60 seconds, so a queue of players keeps moving.
It uses your webcam with Google's MediaPipe face tracking and runs fully offline — no internet, no browser, no servers.
| Title | Duel (best-of-3) | Session stats |
|---|---|---|
 |
 |
 |
A native window, reads the webcam directly. No browser involved.
cd neck_excer
./play.shThe first run creates a local Python environment and installs the dependencies (one time, ~1 minute). After that it launches instantly. To run it manually:
.venv/bin/python game_neck.pyOn first launch macOS will ask to let your Terminal use the camera — click Allow. (System Settings → Privacy & Security → Camera, if you need to enable it later.)
| Key | Action |
|---|---|
| 1 | Solo mode |
| 2 | Duel mode (2 players, split screen) |
| SPACE / ENTER | Start · Play again |
| S | Session stats screen (toggle) |
| R | Reset session stats (on the stats screen) |
| F | Toggle fullscreen |
| M | Mute |
| ESC / Q | Quit |
-
Solo — one player, full screen.
-
Duel — two players stand side-by-side in front of the same webcam. The screen splits in two: the player on the left controls the left lantern (cyan), the player on the right controls the right lantern (orange). Each tilts their own head; the game tracks both faces at once and assigns each to its side automatically.
Best of 3: each round is 60 seconds; the higher score wins the round. First player to win 2 rounds wins the match. The match tally shows as pips under each score during play, and a between-rounds screen lets you press SPACE for the next round. A tied round is replayed without awarding a point.
Great for a stall: two friends duel a best-of-3, a small crowd gathers, and both get the neck-mobility movement at the same time.
(Match length is
WINS_NEEDEDnear the top ofgame_neck.py— set to2for best-of-3,3for best-of-5.)
- Press SPACE → a 2-second calibration captures your upright posture, then a 3 · 2 · 1 · GO! countdown with a rising beep on each digit and a bright tone on GO (tilt your head during it to test your lantern — the timer only starts on GO). Press M to mute.
- Tilt your head ear-toward-shoulder to move the lantern left/right.
- Catch 🪔🎭📜🎶🏛️🔭🎨 culture icons. Chain catches for a rising combo — each catch plays a pluck that climbs in pitch as your combo grows.
- ✨ sparks are rare bonus points (sparkle chime). 🌀💀🕳️ glitch-voids scramble your combo (buzzy thunk) — dodge them. Press M to mute all audio.
A soft ambient pad plays gently under each round (a seamless drone loop) and fades out between rounds, so the stall has a calm, installation-like atmosphere. 5. After 60s you get your score, neck range (°), best combo, and full-tilt count.
The on-screen tilt meter and neck range stat quietly turn the exercise into the scoreboard — players naturally reach their full comfortable range.
When the title screen sits idle with no face for ~6 seconds, the game starts a self-running AI demo — a lantern auto-catches icons while "STEP UP — PRESS SPACE TO PLAY" pulses on screen. The moment someone steps in front of the camera (a face is detected) it snaps back to the title, ready to play. Great for drawing a crowd between players. The demo is silent so the stall isn't noisy.
Press S (from the title or any results screen) to see the day's totals:
plays, threads woven, total tilts, average neck range, duel matches, best
score, and the top-5 scores. The numbers accumulate across the whole session
and are saved to .tot_session.json, so they survive an app restart and
auto-reset when the calendar date changes (a fresh tally each day). Press
R on the stats screen to reset manually. A nice number to read out at the
end of the day: total tilts = how much collective neck mobility the stall did. 🧘
- Press F for fullscreen so it looks like an installation.
- Even, front lighting on the player's face = rock-solid tracking.
- Sit/stand ~50–80 cm from the camera. Calibration adapts to each player.
- The little "YOU" panel (bottom-right) shows the live face mesh — part of the wow.
- If a player's left/right feels reversed, set
TILT_SIGN = -1near the top ofgame_neck.py.
GOOD— the collectible emoji + names (swap in motifs for your theme).RARE,BAD,LANTERN— bonus, hazard, and player icons.ROUND_SECONDS— round length (default 60).FULL_TILT_DEG— how far to tilt to reach the edge (default 20°; lower = easier).SMOOTH— tilt responsiveness.
- "could not open the webcam" → another app (Zoom / Teams / FaceTime / Photo
Booth / a browser tab) is holding the camera. Quit it. If still stuck:
sudo killall VDCAssistant AppleCameraAssistant, or reboot. - Tracking not finding your face → improve lighting; face the camera; move closer.
- It falls back to OpenCV eye-tracking automatically if MediaPipe is unavailable.
game_neck.py— the desktop game (this is the app you want)play.sh— one-command launcher (sets up the venv on first run)requirements.txt— Python dependenciesvendor/face_landmarker.task— the offline face modelindex.html/style.css/game.js— the browser edition (full parity; hosted as the live demo)serve.py/start.sh— optional local server for running the browser edition offline
This is light, playful mobility movement, not physiotherapy. It encourages slow, comfortable lateral neck tilts. Players should move gently, stop if anything hurts, and never force range. 🌱
- macOS / Linux / Windows with a webcam and Python 3.9–3.13.
- Internet only for the one-time dependency install; the game itself runs offline.
Built as a FACTS-H Lab installation for a culture × humanities × technology conference. Face tracking by Google MediaPipe; graphics via OpenCV; all sound synthesized at runtime (no audio files).
Released under the MIT License.