AI Facial Expression Classifier

Classify visible facial expressions (happy, sad, angry, surprised, neutral, fear, disgust) with confidence scores.

AI Private · runs on your device Free
Runs locally in your browser AI processing happens on your device. Your files are not uploaded for inference.
AI model
…
Download size
… ·
Device
Checking browser support…

    Classify the visible expression

    The Facial Expression Classifier first finds each face with MediaPipe, then a ViT image classifier (about 54 MB) scores seven expression classes: happy, sad, angry, surprised, neutral, fear and disgust.

    Important: this is not an emotion meter

    The output describes how a face looks in one image, a model-detected expression, not how the person feels. People can look neutral when happy or smile when sad. This tool must not be used to judge emotions, mental health, honesty or intent.

    Suitable uses

    • AI and computer vision teaching
    • Experiments organising photo archives by expression
    • Game and interactive art prototypes

    Images are processed on your device and never sent anywhere.

    How to use AI Facial Expression Classifier

    1. Upload a photo or start the camera.
    2. Faces are detected and expression scores calculated.
    3. Review the most likely expression and scores for each face.
    4. Treat the results as a model estimate.

    Why use this tool?

    Lets you explore expression classification in a privacy-friendly, transparent way.

    FAQ

    How accurate is it?

    The model reaches about 91% accuracy on a test set of posed faces. Real photos with unusual lighting, angles or ambiguous expressions are often less reliable.

    Does it show someone's real feelings?

    No. It only classifies how the face looks and gives no reliable information about the person's inner state.

    Are my images stored?

    No, analysis happens in your browser and no image is saved.

    Related tools