Computer vision visualizer

Face Mesh Visualizer

Track dense facial landmarks, contours, blendshapes, and 3D-style face geometry in the browser with MediaPipe.

Controls

Input mode
Overlay layers

Face mesh works best when the full face is visible. Contours make eyes, lips, and face oval easier to inspect.

Preview

Upload a face photo, use webcam, or run the sample preview.
Face mesh input
Try face detectionHand landmarks

What this visualizer does

It runs MediaPipe FaceLandmarker and overlays dense facial geometry: tessellation, contours, landmark points, and a face bounds guide.

What MediaPipe detects

  • Face landmarks in normalized x, y, z coordinates.
  • Contours around eyes, eyebrows, lips, and face oval.
  • Optional blendshape scores such as smiles and eye movement.

Common use cases

  • AR face filters
  • Expression analysis
  • Avatar control
  • Face alignment
  • Virtual try-on

Code examples

import mediapipe as mp

BaseOptions = mp.tasks.BaseOptions
VisionRunningMode = mp.tasks.vision.RunningMode
FaceLandmarker = mp.tasks.vision.FaceLandmarker
FaceLandmarkerOptions = mp.tasks.vision.FaceLandmarkerOptions

options = FaceLandmarkerOptions(
    base_options=BaseOptions(model_asset_path="face_landmarker.task"),
    running_mode=VisionRunningMode.IMAGE,
    num_faces=1,
    output_face_blendshapes=True
)

with FaceLandmarker.create_from_options(options) as landmarker:
    result = landmarker.detect(mp.Image.create_from_file("face.jpg"))
    print(len(result.face_landmarks[0]))

How the model works

  1. A face detector proposes face regions in the frame.
  2. The landmark model predicts dense normalized 3D points.
  3. Connections turn those points into mesh and contour overlays.
  4. Blendshape heads estimate expression-like facial movements.

Output explanation

Landmark x and y values are normalized to image width and height. The z value estimates depth relative to the face and is useful for 3D-style effects.

FAQ

For best results, keep your face centered, avoid heavy occlusion, and use face detection first when you only need boxes instead of detailed facial geometry.