Loading page
About this tool
Upload two images to execute a face comparison and calculate their similarity index.
AI Face Match compares two photographs by running face-api.js in your browser — a tiny face detector, a 68-point landmark model and a face-recognition network that turns each face into a 128-number descriptor — then measures the Euclidean distance between those two descriptors and maps it to a similarity percentage. A distance of 0 reads as a perfect match and a distance of 1.4 as no resemblance, with the score clamped between 12% and 99%. It is built for look-alike and family-resemblance curiosity, not identity verification, and the images never leave the page.
Open AI Face Match on AltFTool — it loads instantly in your browser.
Click Select Photo 1 and Select Photo 2 to load the First Subject Photo and Second Subject Photo, each under 10MB, and optionally fill the Subject 1 Name and Subject 2 Name boxes that label the report
Press Run AI Facial Comparison — it stays disabled until both photos are in place — and 'Running Neural Feature Matching...' shows while face-api measures the Euclidean distance between the two face descriptors in your browser
Read the similarity percentage in the dial and its verdict, then the Symmetry Breakdown rows Eyes & Brows Symmetry, Nose Bridge & Width Alignment and Jawline & Lips Contour Similarity; Download saves the report as ai-face-match-report.txt, Copy Report puts the same text on the clipboard, and Reset clears both photos and names
The headline percentage comes from the Euclidean distance between two 128-dimension face embeddings, the same metric face recognition libraries use for matching.
The detector, landmark and recognition networks are fetched as static model files and executed in the tab, so neither photo is uploaded anywhere.
Scores are grouped at 90%+ (same person or identical twins), 75–89% (look-alikes), 50–74% (family resemblance) and below 50% (distinct faces).
By Euclidean distance between two 128-value face descriptors, converted with the formula (1 − distance ÷ 1.4) × 100 and clamped to the 12–99% range. Identical images give a near-zero distance and score at the top of the scale; unrelated faces typically produce distances near or above 1.0.
Any format your browser can display, up to 10 MB per photo. Use a front-facing, well-lit image with the whole face visible — heavy shadow, extreme angles, sunglasses or masks can stop the detector finding a face at all.
The tool falls back to a deterministic signature derived from the two file names and sizes, which produces a placeholder score in the 54–84% range rather than a real facial comparison. If the result looks arbitrary, re-upload with clearer, forward-facing photos so the recognition network can actually run.
No. It is an entertainment and curiosity tool: lighting, age, pose, camera lens and photo compression all move the distance score, and a single embedding comparison is not evidence of who someone is. Never use the result for access control, background checks, or any decision that affects a real person.