Tell a live face from a photo of one
Scores a single frame for signs of a presentation attack — a printed photo, a screen replay or a flat mask — and returns the score with the threshold it was compared against. Passive and single-frame: no challenge, no video, no certification.
- Catches printed photos, phone and monitor replays, and flat masks
- Returns the live score and the threshold, not just a verdict
- Refuses to score a frame too blurry, dark or distant rather than guessing
- Says in the response what it does not establish — identity, 3D masks, injected frames
Free plan included — no card needed to try it.
Try it here
Upload a face, or capture one with your camera. Both run the same single-frame check.
No account needed — 3 free runs every 2 days.
Drop a photo of a face
or click to browse · JPEG, PNG, WebP · max 25MB
Result
Upload or capture a face and the result appears here, with the score and the threshold it was compared against.
Common questions
What does this actually detect?
Presentation attacks: a printed photograph held up to the lens, a face replayed on a phone or monitor, and flat masks. It scores one frame and returns the live probability alongside the threshold that frame was compared against, so you can see the number behind the verdict rather than only the verdict.
Can I use this for KYC or identity verification?
Not on its own, and not where a regulator is involved. This is anti-spoofing, not identity verification — it says nothing about who the person is. If you need that, it belongs alongside face similarity against a document photo, and even then the decision is yours to make, not ours. There is also no iBeta or ISO 30107-3 evaluation behind this model. If your compliance requirement names a certification level, this does not meet it, and we would rather say so here than in a procurement call.
Does using the camera make the check more secure than uploading?
No. A captured frame runs the identical single-frame check an uploaded file does — same model, same threshold, same result. Capturing makes it harder to submit an old file, which is worth something, but it does not make the check harder to fool: a phone screen held to the lens defeats both equally, and a virtual camera bypasses the capture step entirely. We offer both because one is more convenient, not because one is safer.
What is the difference between passive and active liveness?
Passive liveness looks at an image and asks whether it came from a real face. Active, or challenge–response, liveness asks the person to do something — blink, turn, move toward the camera — and verifies the action happened. Active is what the identity-verification industry usually means by “liveness”, and it is harder to defeat with a static artefact. This product is passive. It is a screen, not a gate.
Why was a real face scored as a spoof?
The threshold is 0.82 by default, and it is the most consequential number in the service: it trades a real person being told they are a photograph against a photograph being admitted as a person. Harsh backlighting, a heavy screen glare, a very low-light webcam or an unusually flat image can all push a genuine face under it. The response always includes the score, so a borderline result is visible as borderline rather than arriving as a bare rejection.
Why does it sometimes refuse to score instead of answering?
Because a confident number computed from an unusable frame is worse than no number. The model will happily score a 20-pixel blur, and something downstream would act on it, so frames that are too blurry, too dark, too bright, or where the face is too small or too close are returned as low_quality with the reason, and frames with no face as no_face. Neither is a pass. Nothing in this service resolves to “live” by default — when the models cannot be reached at all it answers 503 rather than guessing.
Does it work on video?
Not as video. You can pass any single frame extracted from one, which is what the camera mode here does, but there is no tracking across frames and no motion analysis. Each call sees one image and nothing else.
What happens to the image I submit?
The models run on our own servers — nothing is sent to a third-party AI provider, and nothing you submit trains any model, including ours. Signed in, a run is kept so you can return to it and is deleted when you delete it or your account. The privacy policy sets out the retention in full. You get 3 runs here without an account.