Skip to content

Use case · Deepfake and spoof defence

Deepfake and spoof defence

A face on a screenis nota person.

A selfie check is only as good as its answer to a photo held up to the camera, a video on a second phone, or a synthetic face fed straight into the camera stream. Toden layers a randomised challenge, a server-side liveness model, a check for a phone in view and device integrity signals, and sends anything doubtful to a person.

The workflow

Layered liveness: what the customer is asked to do, and what our servers see in the recording. This is a starting point: add, remove or reorder any step in the workflow builder.

Step 01

Document Intelligence

Passport, licence or national ID

Per check

Step 02

NFC Chip Verification

Proves the passport chip is genuine

Per check

Step 03

3D Active Motion

Random head turn and blink prompts

Per check

Step 04

3D Flash Check

A random colour sequence reflected off the face

Per check

Step 05

Facial Match

Selfie against the chip portrait

Per check

Step 06

Device Intelligence

Virtual cameras, emulators, app integrity

Per check

Then your rules decide

A starting rule set

  • The liveness model is confident it is a photo or a screenDecline
  • A phone is visible in front of the cameraDecline
  • The request did not come from your registered appDecline
  • A virtual camera or an emulator is detectedReview
  • No recording reached usReview
  • A live person on a genuine deviceApprove

Every rule is yours to change on a visual canvas, and each result records which rule decided it. Publishing a new version needs no release.

Try the workflow builder

How it works

1

Ask for something unpredictable

Gestures in a random order, or a random sequence of colours flashed from the screen, which a recording made yesterday cannot answer.

2

Judge the recording on the server

A liveness model scores the selfie and the video, a second model looks for a phone held in front of the camera, and the colours are checked against the face.

3

Check the device

Virtual cameras, emulators, rooted phones and apps that are not yours are reported, and your workflow decides what each one means.

What powers it

  1. The liveness verdict comes from our servers reading the recording, so a modified app cannot simply report a pass.

Integrate

Mount the SDK with the workflow you published, or send a hosted link and write no code. The result arrives by signed webhook, and changing the workflow never needs a release.

Verify.tsxTypeScript · React
import { TodenVerification } from "@todenhq/sdk-react";
export function Verify({ userId }) {
return (
<TodenVerification
apiKey="pk_test_..."
workflowId="wf_AbC123dEf456"
userId={userId}
onSubmit={(s) => save(s.verificationId)}
/>
);
}
verification.completedWebhook · JSON
{
"event": "verification.completed",
"data": {
"verificationId": "ver_8fk2...",
"status": "verified",
"assuranceLevel": "gov_db",
"idType": "bvn",
"country": "NG",
"dataMatch": true,
"facialMatch": true,
"facialConfidence": 91
}
}

Frequently asked

  • No single check stops every attack, and anyone who says theirs does should be asked for the evidence. Layers make an attack get past several independent checks at once, and a doubtful capture goes to review.

Deepfake and spoof defence,in one workflow.

The sandbox opens when you sign up. Usage-based pricing, no sales call.