Skip to main content
Identify AI face-swapping, synthetic overlays, and deepfake masks in real time as applicants complete identity verification — adding a critical layer of defence to your onboarding process.

​
What is Live Deepfake Detection?

Live Deepfake Detection is a dedicated workflow step that analyses both the video frames and audio captured during an identity verification session. It runs three independent analysis layers in parallel — liveness verification, vision-based deepfake analysis, and voice verification — to produce a final verdict of LIVE, DEEPFAKE, or INCONCLUSIVE. The system captures 6 frames and an audio clip during the session. The applicant is presented with a challenge phrase (3 random words) that they must speak aloud, enabling the voice verification layer to detect synthetic speech and text-to-speech attacks alongside the visual analysis. While Face Liveness confirms a real, physically present person is in front of the camera, Deepfake Detection goes further — it determines whether the face being presented is authentic or AI-generated, and whether the voice is a real human voice or a synthetic reproduction. Together, these layers provide comprehensive protection against face-swap deepfakes, presentation attacks, and synthetic voice attacks.
1

Session initiated

The applicant begins their identity verification session through the hosted verification link or embedded flow.
2

Frame and audio capture

The applicant’s device captures 6 video frames at key moments — a still frame, a blink, left and right head turns, and a speaking frame — along with an audio clip of the applicant reading the challenge phrase aloud. All data is transmitted securely to deepidv’s analysis infrastructure.
3

Three-layer parallel analysis

Three independent analysis layers run simultaneously for a total processing time of 5–7 seconds:
  • Liveness verification — validates facial landmark data across all frames, confirming real blinks, head turns, and proper frame timing to rule out photo/video replay attacks
  • Vision analysis — AI-powered inspection of the captured frames for face-swap artifacts including boundary seams, texture smoothing, unnatural teeth rendering, flat rotation without depth, temporal instability across frames, and skin tone mismatches
  • Voice verification — analyses the audio for biological sounds (breathing, lip smacks), natural voice onset/offset patterns, consonant variation, pitch micro-tremors, and replay artifacts to detect text-to-speech and voice cloning attacks
4

Verdict determination

A verdict engine combines the results from all three layers into a final determination: LIVE (authentic), DEEPFAKE (synthetic content detected), or INCONCLUSIVE (insufficient confidence to make a definitive call). A combined risk_score and confidence value accompany the verdict.
5

Results recorded

Detection results are included in the session’s analysis_data object as deepfake_detection_data, available via the Retrieve Session endpoint, webhook events, and visible in the Admin Console.

​
Enabling Deepfake Detection

​
Via the Admin Console

  1. Navigate to Workflows in the sidebar
  2. Select an existing workflow or click Create New to build one
  3. Locate the Deepfake Detection toggle in the workflow configuration
  4. Enable the toggle to add deepfake detection to the workflow
  5. Click Save to apply changes
All future sessions created with this workflow will automatically include deepfake detection analysis.

​
Via the API

When creating a session programmatically, reference a workflowId that has Deepfake Detection enabled: 
curl -X POST https://api.deepidv.com/v1/sessions \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_API_KEY" \
  -d '{
    "firstName": "Jane",
    "lastName": "Smith",
    "email": "jane.smith@example.com",
    "phone": "+14165557890",
    "workflowId": "your-deepfake-enabled-workflow-id"
  }'
Find your workflowId under Workflows in the Admin Console sidebar. Ensure the workflow has the Deepfake Detection toggle enabled before creating sessions.
For maximum protection, combine Deepfake Detection with Face Liveness and ID Verification in a single workflow. This layered approach addresses presentation attacks, digital injection, and document fraud simultaneously.

Interpreting Results

Understand deepfake detection scores, signals, and confidence thresholds.

Workflows

Learn how to build and manage verification workflows with deepfake detection.