Loading demo…
Detection & attributesLive
Detect faces in an image — free
Upload a photo and get every face in it back in one call: bounding boxes, facial landmarks, and age and gender estimates with per-face confidence scores. No account required.
- Multi-face detection in a single call
- Bounding boxes + 5-point landmarks
- Age & gender estimation per face
- ~80ms median latency
Live demo
Free face detection tool
Upload a photo with one or more faces to see detected bounding boxes and per-face attributes.
Input image
Detected faces
Upload an image to detect faces
How it works
How face detection works
Detection is the foundation of every other face task. Before you can compare two faces, recognise one, or blur a crowd, something has to find the faces and say where they are. That is all this step does — it locates faces without identifying anybody.
- 1
Upload an image
Any photograph with one or more faces. The detector works on the full frame, so you do not need to crop or point it at a region first.
- 2
Faces are located
A single forward pass returns a bounding box for every face in the image, each with a detection confidence score. Group shots resolve in the same call as a portrait.
- 3
Landmarks are extracted
Five key points — both eyes, the nose tip, and both mouth corners — are placed on each face. These drive the alignment step that comparison and recognition depend on.
- 4
Attributes are estimated
Age and gender estimates are produced per face from appearance alone. They are returned alongside the geometry so a single call gives you everything about the frame.
Output
What you get back per face
Every detected face comes back as a structured record. The demo renders these as overlays; the API returns them as JSON.
| Field | Type | Meaning |
|---|---|---|
| bbox | [x1, y1, x2, y2] | Pixel coordinates of the face rectangle, in the source image coordinate space. |
| det_score | number | Detection confidence from 0 to 1. Low values usually mean a small, occluded, or ambiguous face. |
| landmarks | [[x, y], …] | Five points: left eye, right eye, nose tip, left mouth corner, right mouth corner. |
| age | number | Estimated age in years. A statistical estimate with several years of typical error. |
| gender | string | Predicted gender from facial appearance. An inference, not a stated identity. |
| image_width / image_height | number | Source dimensions, so boxes and landmarks can be scaled onto your own canvas. |
Age and gender are inferences drawn from appearance, and they carry real error. Using them to make consequential decisions about an individual is both unreliable and, in several jurisdictions, legally fraught — see GDPR and biometric data for what applies when you process facial images in the EU.
Image quality
Which images work best
Detection is robust, but it is not magic. The main variable is how many pixels the face actually occupies.
Detects reliably
- Faces at least roughly 40×40 pixels in the source
- Frontal and three-quarter views
- Multiple faces at varying depths in one frame
- Ordinary indoor or outdoor lighting
- Standard JPEG, PNG, or WebP input
Harder to detect
- Very distant faces in wide landscape shots
- Full profile or back-of-head views
- Strong backlighting that silhouettes the face
- Significant motion blur
- Illustrations, avatars, and heavily stylised art
Use cases
What teams use face detection for
Photo pipeline preprocessing
Find and crop faces before comparison, recognition, or blurring — detection is the step every other face task builds on.
Content moderation
Flag images containing people, count how many, and route them for review based on what is actually in the frame.
Audience analytics
Estimate aggregate age and gender distribution across a set of images, for retail or media measurement.
Capture quality gating
Reject a selfie upload before it reaches your KYC flow when no face is found, several are, or confidence is too low.
Demo vs API
From free tool to production API
The demo draws the boxes for you. The API hands you the coordinates so you can draw them yourself — or feed them straight into the next stage of your pipeline.
| Free demo | API | |
|---|---|---|
| Model & detection rate | Identical | Identical |
| Access | Browser upload | REST, multipart in / JSON out |
| Output | Annotated image + attribute table | Structured face array with coordinates |
| Volume | A few images | Rate limits by plan |
| Uptime SLA | Best effort | Contractual |
| DPA & compliance docs | Not applicable | Included on enterprise |
FAQ
Face detection FAQ
- Is the face detection tool free?
- Yes. The live demo is completely free — no account, credit card, or signup required. Upload an image and see every detected face with its attributes.
- How do I detect faces in an image online?
- Upload a photo using the demo above. The detector scans the full frame and returns a bounding box for every face it finds, along with five facial landmarks and estimated age and gender for each one.
- What is the difference between face detection and face recognition?
- Detection answers "is there a face here, and where?" — it locates faces without identifying anyone. Recognition answers "whose face is this?" by matching against a known identity. Detection is the first step in almost every face pipeline, including recognition, comparison, and blurring.
- How many faces can it detect in one image?
- There is no fixed cap. The detector processes the entire frame in a single pass, so group photos and crowd scenes are handled in one call. Detection reliability drops as individual faces get very small relative to the image.
- What is the detection score?
- Each face is returned with a detection confidence between 0 and 1, indicating how certain the model is that the region contains a face. Low-scoring detections in cluttered scenes are worth reviewing before you act on them automatically.
- How accurate are the age and gender estimates?
- They are statistical estimates from facial appearance, not ground truth. Age in particular carries meaningful error, typically several years, and accuracy degrades with poor lighting, extreme pose, and heavy makeup or filters. Treat them as aggregate signals rather than facts about an individual.
- What are the five landmarks?
- The two eye centres, the nose tip, and the two mouth corners. These points are what downstream steps use to align a face to a canonical pose before comparison or recognition, which is why alignment-sensitive tasks depend on detection quality.
- Can I use this as a face detection API?
- Yes. The same model powering this free demo is available via REST API. It returns a JSON array of faces with bounding boxes, detection scores, landmarks, and attributes, plus the source image dimensions so coordinates can be mapped back onto your own rendering. Enterprise plans include SLAs and a Data Processing Agreement.
Related reading
Go deeper
Understanding face recognition accuracy
What TAR and FAR actually mean, and why a single accuracy percentage tells you almost nothing.
ReadGDPR and biometric data
What Article 9 special category data means in practice when you are handling facial images.
ReadRelated tools: face comparison, face blur, and liveness detection.
Ready to ship identity verification?
Talk to our engineering team about your use case. We'll get you up and running in under a day.
SOC 2 Type II · GDPR Ready · ISO 27001