FACIAL ACTION CODING SYSTEM
Calibrate your face. Isolate an action. Build a combination.
Scores are calibrated movement estimates. They are not formal FACS grades.
Raise the target slider to hold a pose. Lower it to zero between repetitions.
SMALL MOVEMENTS, BUILDING BLOCKS
Grounded in facial movement research.
Informed by Melinda Ozel’s public FACS resources. Original exercises; no affiliation or endorsement.
A VOCABULARY FOR MOVEMENT
23 actions with camera estimates. 9 actions for unscored mirror practice.
Use the original photo and motion references to check the pose. Our drawings are schematic guides.
No matching actions. Try an AU number or another region.
YOUR OWN POINT OF COMPARISON
Completed exercises stay in this browser. Calibration is kept only for the current camera session.
WHAT THE FEEDBACK MEANS
A practice aid for learning facial control, informed by public FACS references.
Capture your relaxed face in steady light. Before scoring an action, record a comfortable example and check it against the linked reference. Accept it only if the movement matches the cue.
Low or noisy ranges remain unscored. Recalibrate when the lighting or camera position changes. Calibration cannot correct every model error.
The score compares changes from neutral with your accepted range. For two-sided targets, the weaker normalized side controls the result. Isolation is a separate requirement: either side of a selected competing action can interrupt a hold. Quiet-region checks compare changes with your neutral face and allow a small noise-dependent tolerance.
A full brow raise (AU1 + AU2) cannot pass an isolated outer brow raiser (AU2) challenge. Calibration rejects sustained extra movement too. A pass requires three holds with releases between them. Blinks require a brief closure and reopening. Scores are neither AU probabilities nor the FACS A–E intensity scale. A high target match alone cannot pass an isolation check. These checks estimate visible movement, not which muscle fibers are contracting.
Complete the individual components to unlock a combination. The coach checks every component together. If one component drops out, the hold restarts.
Challenges are named for their movements. Facial actions alone do not determine emotion or sincerity. The sequence and scoring thresholds are original design choices, not an Ozel-endorsed course.
MediaPipe estimates landmarks and blendshapes, not muscle activity. AU6/7 and nearby lip actions can be confused. Jaw opening does not tell the model whether the jaw was relaxed or actively stretched.
AU8, 11, 13, 21, 23, 27, 31, 38, and 39 are offered as unscored mirror exercises. These 32 entries are a practice library, not the entire FACS coding system. Detection accuracy has not been validated against certified coders.
PRIVATE, BY CONSTRUCTION
The bundled tracker runs locally in your browser. The app does not record, upload, identify, or store your face. It requests video only; no microphone permission.
Only exercise IDs, completion counts, best practice scores, and completion dates are saved locally. Calibration and camera frames remain in memory and are discarded when the camera stops or the page closes. Preview sessions save nothing. External reference links open their original websites when clicked.
READ THE ORIGINALS
Guide-arrow design informed by FacePunk’s public demo videos ↗. The arrows and quiet-region markers here are original drawings.
Tracker: Google MediaPipe Face Landmarker, Tasks Vision 0.10.21. Documentation ↗ · License ↗