stile

Liveness detection

Liveness detection

Facial liveness detection determines whether a face presented to a camera comes from a live, physically present person rather than a photo, mask, replay, or deepfake. In practice, it stops someone from passing a face check by holding up another person's picture.

  • Liveness technology tested by iBeta, Level 1 and 2
  • Passive by default
Illustration: a phone showing a completed liveness check.

How a liveness check works

For the person, it feels like taking a short selfie video. Behind the scenes, it happens in four steps.

4 scenarios · Scroll to explore
Illustration: a woman holds her phone up at arm's length to take a selfie for a liveness check.

The person looks at the camera

In this step
A prompt to fit their face inside an oval, on their phone or computer.
Illustration: a phone running a liveness check, with the person's face inside an oval and the prompt Look straight at the camera.

The camera records a short clip

In this step
Hold still for a moment. There is no head turn or phrase to repeat.
Illustration: a printed photo of a man's face beside a silicone mask of an older man, two common presentation attacks.

The clip is checked for fakes

In this step
  • A printed photo
  • A phone or tablet screen
  • A mask
  • A replayed video
Illustration: a laptop shows a webhook with a liveness score, beside a result card reading Liveness check, Live person confirmed: result passed, capture passive, signature verified.

The business gets the answer

In this step
  • Real person or not
  • When the check ran
  • A signature proving it came from Stile

What it stops

Fake faces shown to the camera

Most attacks are simple: someone shows the camera a photo, a video or a mask of another person. These are called presentation attacks, and stopping them is the job of liveness detection.

  • Stops

    Presentation attacks

    Photos, screens, masks and replayed videos shown to the camera.

  • Doesn't cover

    Injection attacks

    Video fed straight into the device by software, skipping the camera. Many deepfakes work this way. Stile doesn't claim to detect deepfakes.

Illustration: on the left, a hand holds a printed photo up to a phone running a liveness check. On the right, a laptop plays a video of the same face into a phone through a cable, so it never passes in front of the camera.
Lab testing

Results for the liveness technology Stile uses, from iBeta's confirmation letters.

In independent lab testing, no attack got through

  • Level 1 · 11 September 20230 of 900attacks got through6 kinds of attack, 150 attempts each
  • Level 2 · 19 October 20230 of 750attacks got through5 harder kinds of attack, 150 attempts each

What the levels mean. Level 1 attacks take up to 8 hours to prepare. Level 2 attacks take up to 24 hours and more skill.

What was tested. The same liveness technology Stile runs in every liveness check, tested on a Samsung Galaxy S21 (Android 12).

Active vs passive

Active, passive and hybrid liveness

Liveness checks differ in what they ask the person to do. That is a choice about the experience, not about security.

  • Stile's default

    Passive

    The person just looks at the camera.

  • Prompted

    Active

    The person follows a prompt, such as turning their head.

  • Screen-led

    Hybrid

    The screen does something, such as flashing a light, while the person looks.

Lab tests find strong and weak products in every group, so compare tested results, not labels.

Liveness detection vs face match vs face recognition

These three are often confused. Each one answers a different question.

  • Liveness detection

    Is a real person in front of the camera right now? It doesn't say who they are.

  • Face match

    Is this the same person as the photo on their ID? One face is compared with one photo. How biometric verification works.

  • Face recognition

    Who is this? One face is searched against many. It is used to identify people, not to verify them.

Choosing a provider

Four questions to ask a liveness provider

  • Level

    Which iBeta levels has it passed?

    Level 2 covers more skilled attacks than Level 1. Ask for both.

  • Date

    When was it tested?

    iBeta publishes a dated letter for each result. Read it.

  • Setup

    What exactly was tested?

    The letter names the device and settings. Compare them with how you will use it.

  • Gaps

    What wasn't tested?

    Lab tests cover fakes shown to the camera, not video fed in by software. Ask how that is handled.

For developers

Add liveness detection to your app

Your server starts a check, the person completes it on their own device, and your server receives the result. Every result is signed, so you can confirm it came from Stile.

Stile doesn't send the video or photos. Raw ID images and biometric captures are deleted on completion by default; signed outcomes and hashed anchors may persist for audit.

Example result, shortenedjson
{  "id": "evt_example",  "object": "event",  "type": "verification_session.verified",  "data": {    "object": {      "id": "vks_example",      "object": "verification_session",      "status": "verified",      "client_reference_id": "order_123",      "verification_result": {        "liveness_score": 0.97      }    }  },  "created": 1789992000}

FAQ

Common questions

Still have a question? Book a demo

A liveness check, sometimes called a liveness test, makes sure the face in front of the camera belongs to a real person who is there at that moment. For the person, it feels like taking a short selfie video. Liveness detection is the technology that makes the decision.

Neither, by default. A passive check asks the person to do nothing; an active check asks for a movement, such as turning their head. Lab results vary widely within both groups, so look at how a specific product was tested. Stile uses passive checks by default.

Not on its own. Liveness detection is tested against fakes shown to the camera, such as photos, screens and masks. A deepfake fed straight into the device by software never passes in front of the camera, and liveness lab tests don't cover it. Stile doesn't claim to detect deepfakes.

Liveness detection checks that a real person is there. Face match checks that they are the person on the ID. Identity checks usually need both. How biometric verification works.

Yes. The liveness technology Stile uses was tested by iBeta, an accredited lab, against the ISO/IEC 30107-3 standard at Level 1 and Level 2, and no attack got through. The confirmation letters are dated 11 September and 19 October 2023. What the iBeta levels mean.

A liveness check on its own costs $0.03. Every account gets 1,000 free checks a month. See pricing.

Yes. You can run a liveness check by itself, or add an ID check and face match when you also need to know who the person is.

Passive. The person just looks at the camera. You can switch to active or hybrid if your flow allows a prompt.

Whether the check passed, the status of the session and when it happened, signed so you can confirm it came from Stile. It doesn't send the video, the photos or a face template.

Stile uses it to run the check. By default, raw ID images and biometric captures are deleted on completion; signed outcomes and hashed anchors may persist for audit. How Stile handles data.

Your developers start a check from your server, show it to the person in your app or website, and receive the result. Read the integration guide.

Get started

Try liveness detection for free

1,000 free checks every month. No credit card needed.