Loading page
About this tool
Identify and analyze human facial expressions and emotions in real-time or from uploaded photos using AI.
The Emotion Detector runs face-api.js models in your browser to find faces in a photo or webcam frame and score each one across seven expression classes — happy, sad, angry, surprised, fearful, disgusted and neutral — returning a confidence value for every class rather than a single label. It uses the Tiny Face Detector at a 512px input size with a 0.5 score threshold, plus the 68-point facial landmark model to trace jaw, brows, eyes, nose and mouth. Models are downloaded once and everything after that runs locally with WebGL, so no image is ever uploaded.
Open Emotion Detector on AltFTool — it loads instantly in your browser.
Choose Upload Photo or Live Webcam, then drop a JPG, PNG or WEBP under 10MB onto the Select Photo area.
Wait while "Running AI Face Analytics..." shows — the Tiny Face Detector runs at inputSize 512 with scoreThreshold 0.5 and traces 68 facial landmarks.
Read the Analysis Report: Dominant Mood, an Emotion Matrix bar for all seven expressions, and Technical Integrity scored out of 100, then hit Download, Share or Copy.
Every detected face returns a score for all seven expressions, so you can see a 0.61 happy alongside a 0.28 neutral instead of just the winning label.
Each face gets a 0-100 quality read built from average brightness, edge strength and detection score, surfaced as lighting, blur and visibility ratings so you know when a low confidence is the photo's fault.
detectAllFaces returns every face above the threshold, each with its own bounding box, landmarks and emotion matrix.
Seven: happy, sad, angry, surprised, fearful, disgusted and neutral. These are the classes the face-api.js expression network was trained on, and each face gets a confidence score for all seven that sum to roughly 1.
No. The detection and landmark models are fetched once from this site and then run entirely in your browser tab via WebGL; the image data itself never leaves your device. Recent results are kept in sessionStorage — the last 10 entries — and clear when you close the tab.
The Tiny Face Detector only reports faces scoring above 0.5 confidence at a 512-pixel input size, so small, heavily angled, partially covered or poorly lit faces get dropped. The quality panel flags this: average brightness under 80 is rated Poor lighting, and low edge strength is reported as High blur.
Treat the output as an estimate of facial expression, not of what a person feels. These models infer from muscle geometry alone and are affected by lighting, camera angle, glasses, facial hair and cultural differences in expression — they should never be used to make decisions about individuals in hiring, security, healthcare or education.