gitignore POSTS.md alongside the other local filming aids

gitignore teleprompter.html alongside SCRIPT.md (local filming aids)

Restore eye-sum head_x mapping (squint forward, wide back)

Antenna rewrite: brow_down adds frown, brow_asym kicks single-brow gestures

Use the calibrated −14° pitch bias (rest = 0°)

Calibrate offsets from snapshots; drop eye-diff yaw; +50% jawOpen wobble

Add 'Save snapshot' button: dump full tracking state to JSON

Lock body_yaw to head_yaw (same angle) to fix yaw cross-coupling

Pitch bias -10° → -20° per user feedback (still tipping down at rest)

Flip pitch bias sign +10°→−10°; rest pose was tipping chin down

Disable mouthShrugLower→head_z (commented out, easy to restore)

Add +10° pitch bias to keep resting head off self-collision

Add camera picker dropdown to controls row

Remove SCRIPT.md from repo; keep local only

Add demo video script — ~90 s, beat-by-beat with stage directions

Cut smile→head_z amplitude; remove wink-flick

Smile→head_z, wink→yaw flick, drop nostril_flare from body_yaw

Shift head_x bias +1 cm forward; bump brow→antenna weight 0.9→1.4

Decouple head position from face position; tame eye-X amplitude

Cancel resting eyeSquint baseline so head_x sits at 0 at rest

Move speech wobble off head_x onto head_pitch+roll; double amps

Speech wobble: jawOpen nods pitch down + leans head forward

Bump eye-diff yaw weight 15→90°/feature; add MAPPING.md

Fix axis mix-up: forward/back is X (head[0][3]), not Z

New default: amps ×1.5, eye-sum→z, eye-diff→yaw, mouthPucker→y

Fix desktop grid indices so chart sits flush under webcam

Add AGENTS.md — handoff notes for cross-machine continuation

Fold status row + Stop into the controls panel so chart sits flush under the video

Move blendshape monitor under video pair; add experimental nostril_flare derived feature

Live blendshape monitor: 52 horizontal bars updating at 15 Hz

Side-by-side raw + tracked video with full mesh, eyes, brows, lips, irises

Flip pitch sign — looking down now nods robot down

Bump MediaPipe tasks-vision to 0.10.35 (0.10.22 was a 404 on jsdelivr)

Lazy-load MediaPipe to unblock DOMContentLoaded (was hanging at Checking sign-in)

Smooth startup via Motor.softReturnToBase; clean pagehide/beforeunload shutdown

Auto-connect after authenticate; one-tap robot picker; checking-in state

Vendor reachy-mini.js (same fork marionette_js uses) and import locally

Return to neutral on STOP and master-off; bump tap target sizes for mobile

Tuner UI: head amplitude sliders + free-routing rule editor with live readouts

Map face features → robot DOFs, send setTarget at 20 Hz with safety clamps

Integrate MediaPipe FaceLandmarker; log blendshapes and head pose

Acquire user webcam on session start, stop on session end; mirror toggle

Wire HF auth, signaling connect, robot picker, session lifecycle

Scaffold README, index.html, style.css, main.js stub

Add plan.md — minimal live face-mimicry app

initial commit