...
A fraud analyst reviewing a flagged face check

Deepfake Detection: Generated Faces, Face Swaps and Injected Video

A generated face, a swapped face or a video fed in place of the camera all look like a normal selfie to your onboarding flow. Our deepfake detection software checks the image and the stream in the same session as liveness and the document check, and names the reason when something fails.

See What We FlagWhat we flag
  • Deepfake-resistant Deepfake-resistant
  • Resistant to video stream injection Stream injection
  • Part of the KYC session In the session
  • Named checks in the API response Named checks

12,000 free verifications when you first connect

Every company connecting to NeuroVision for the first time gets 12,000 free verifications for 12 months. Available from December 1, 2026.

Learn more

Independently tested and certified

ISO register global certification markASIB certification markQA Compliance Assessment certification mark
  • ISO/IEC 27001:2022
  • GDPR
  • iBeta ready
Global coverage

16,000+ document types
from 195+ countries and territories

Customers verify with the ID they actually carry — in 90+ languages, including handwritten fields and MRZ.

16,000+

document types


Passport ID card Driver's licence Residence permit Visa Work permit Voter card Tax ID eID Biometric passport Military ID Health card National ID Travel document Refugee ID Social security card
195+

countries and territories


Kazakhstan UAE Germany United Kingdom Singapore India Brazil Nigeria Philippines Vietnam Türkiye Mexico Uzbekistan Egypt Indonesia France Spain Italy Poland Kenya Saudi Arabia Japan South Korea Australia United States Canada Thailand Malaysia South Africa
What we flag

What Our Deepfake Detection Flags

Four ways to fake a face, and two ways to deliver it.

Every check below runs on the customer's selfie, in the same session as liveness and the document check. These are face checks inside identity verification — not a general media detector.

AI-generated faces

Faces produced by generative models — GAN or diffusion — are flagged as signs of generation rather than a photograph of a person.

GAN · Diffusion

Face swap

A real recording with someone else's face mapped onto it is flagged, which is the form most onboarding attacks take.

Swapped face

Fully generated images

An image that was generated end to end, rather than captured by a camera, is flagged even when it looks like an ordinary selfie.

Generated image

Deepfake video shown to the camera

A deepfake played back on another screen in front of the camera is caught as a presentation attack, alongside replays, masks and 3D heads.

Presentation attack

Injected video streams

A prepared video fed in place of the camera never reaches the lens at all. Our liveness is resistant to video stream injection.

Stream injection

Someone else's face in the frame

A second face in the frame is detected, so a customer cannot be prompted or stood in for by another person during the check.

Foreign face
In the flow

Deepfake Detection Inside Your KYC Session

There is no separate deepfake product to integrate. The checks are steps of the session you already run.

Selfie, liveness and combined scenarios are configured in the scenario editor, not in your code.

  1. 1. Capture through our widget

    The customer's face is captured by our widget in the browser, not uploaded as a file they prepared earlier.

  2. 2. Confirm a live person

    The liveness step confirms a real person is in front of the camera and rejects printouts, screens, masks and 3D heads.

  3. 3. Check the image itself

    The selfie is checked for signs of generation and face swap, and for a second face in the frame.

  4. 4. Get a result with a reason

    Each check comes back named, with its own result. A failed face check returns a code and a description your support team can act on.

Entry points

Where a Fake Face Tries to Get In

The same fake face arrives in three different ways, and each one is caught by a different check.

A deepfake played back on a screen

The attacker plays a generated or swapped-face video on another phone, tablet or monitor in front of your customer's camera. This is a presentation attack, and it is caught by the same liveness step that stops printouts, masks and 3D heads.

When During the liveness step
Checks
  • Deepfake video
  • Video replay
  • Screen photo
  • Masks and 3D heads

Deepfake video is also a liveness topic

A deepfake shown to the camera is one of the presentation attacks our liveness detection stops, together with printouts, masks and 3D heads.

Response

What You Get Back

A named check with a confidence value

The face checks come back in the session response as named items — the deepfake check among them — each with its own status and confidence, so you can set your own thresholds on top of ours.

A result you can act on, not a verdict you have to trust.

Every check is logged

Each check, its result and the decision that followed are stored with the session, ready for an audit or a dispute.

The best frame goes on to face matching

The frame chosen during the liveness step is the one compared with the photo in the document, so a failed deepfake check and a failed match are two separate answers.

Borderline cases go to review

A questionable result goes to your team in the dashboard instead of rejecting a real customer outright.

A KYC session record showing the result of each face check

Manual review and analytics are part of the back office, not an extra product.

Integration

Deepfake Detection API

Face checks run on every selfie you send through our REST API or capture with the Web SDK. There is no separate deepfake endpoint to call, and no separate media-analysis service to buy.

View API Documentation

Ways to integrate

Part of the KYC sessionNo separate integration

Turn the selfie and liveness steps on in your scenario, and the face checks run with them.

Named checks in the responseFor your rules engine

The session response carries each face check by name with its own result and confidence.

Sandbox with test dataBefore you go live

Test the flow on test data first, with help from our integration team.

What comes with the integration

  • Session API and webhooks
  • Named face checks with confidence
  • Public error reference
  • Scenario editor
  • Sandbox from day one

Cloud or on-premises

On-premises keeps data in your jurisdiction.

Review in the dashboard

  • Session history
  • Check results
  • Manual review
  • Export
Limits

What We Claim and What We Do Not

No deepfake detection rates

We do not publish detection rates for deepfakes, and we would treat anyone's published number with care: results depend on the attack, the device and the generator, and a figure without those conditions says very little.

One layer, not the whole answer

A face check is one layer. Liveness decides whether a live person is in front of the camera, and the document check decides whether the identity is real. A deepfake that fails none of them still has to match the ID.

Liveness detection

Where the check is strongest

Our checks are strongest when the face is captured through our widget rather than uploaded as a file, because live capture removes the easiest delivery route.

Face verification

Deepfake-resistant, and resistant to video stream injection

Those are the two claims we make. Everything else on this page is a list of what the checks look for.

Industries

Deepfake Detection by Industry

A generated face costs the most where the face is the key to money.

  • Banks and digital banks

    Remote account opening and high-value confirmations both rest on the face being real, not rendered.

    Remote onboarding High-value confirmation
  • E-wallets and payments

    Account upgrades are where limits rise and a generated selfie becomes worth making.

    Account upgrade Higher limits
  • Crypto exchanges

    Traders are verified remotely and paid out quickly, which leaves little time to spot a fake face by hand.

    Onboarding Before withdrawal
  • Digital lending

    A borrower who never existed is the cheapest loan to fake and the hardest to collect.

    At application Before payout
  • Marketplaces and gig platforms

    Sellers, drivers and couriers are onboarded at scale and rarely seen in person.

    Seller onboarding At scale
  • Gaming and betting

    Licensed operators verify players before withdrawals and keep the evidence for the licence audit.

    Before withdrawal Audit trail
Why us

Why teams choose NeuroVision Global Systems

Accuracy you can audit

Independently tested face matching, a confidence score on every result.

Accuracy you can audit

The face algorithm was tested by NIST and placed TOP-24, NIST FRVT 1:1 Verification. Every document and face result comes with a confidence score.

TOP-24NIST FRVT 1:1 Verification
0.1 sper face comparison

Live in under an hour

REST API, Web SDK, iOS and Android SDK, WebView; sandbox first.

Live in under an hour

Under an hour for REST API or Web SDK with sandbox keys; production in one business day. A 14-day trial on your real traffic comes with full support.

< 1 hto integrate with sandbox keys
14 daystrial with full API access

Global coverage

195+ countries and territories, 16,000+ document types, 90+ languages.

Global coverage

Customers verify with the ID they actually carry: 16,000+ document types from 195+ countries and territories, including handwritten fields and MRZ.

16,000+document types
90+languages

Anti-fraud built on real attacks

40+ document checks, 1:N face search, blacklist screening, 30 more layers.

Anti-fraud built on real attacks

Deepfake-resistant liveness, 1:N deduplication and blacklist screening on top of document checks — one person, one account.

40+document checks
0.01%residual fraud risk

Privacy by design

ISO/IEC 27001:2022, GDPR-ready; on-premises, cloud or hybrid.

Privacy by design

Run it in our cloud, in your data centres or hybrid — templates and personal data never leave your jurisdiction.

27001:2022ISO/IEC certified
GDPRready

In-house engines only

OCR, face, liveness and anti-fraud are ours.

In-house engines only

One vendor covers KYC, KYB, AML and KYT on one API — one SLA, no third-party licences.

20 yearsof computer-vision engineering
1 SLAfor every check
Pricing

Deepfake Detection Pricing

Face checks are not billed separately. They come with the package that includes the selfie and liveness steps.

Basic package

From $0.44 per check

The Basic package adds the selfie and liveness steps to the document check. Build your own set in the price calculator.

Calculate your price

Pay as you go

No subscription and no separate contract for face checks.

See Pricing

Free trial

A 14-day trial on your real traffic with full API access and sandbox access from day one.

Volume and enterprise

Custom terms for high volumes, on-premises deployment and an SLA.

Products

Our AI-Powered Identity Solutions

Everything you need to verify, authenticate, and protect your customers — all in one powerful platform.

A real person with a real document — verified in seconds. Document capture, face match, liveness, anti-fraud and AML.

Learn more

Screen every customer against sanctions, PEP and watchlists, including self-exclusion lists — inside the same verification flow.

Learn more

Verify the business behind the account: company registry data and ultimate beneficial owners (UBO), checked before onboarding.

Learn more

Transaction monitoring in real time: suspicious transfers are flagged for review, on the same API as KYC and AML.

Learn more

A real, present person — deepfake-resistant. Authenticate returning customers by face instead of passwords.

Learn more

Payments confirmed with a face or an ID document — the identity your customer already verified, at checkout.

Learn more
Contacts

Try Deepfake Detection Free

Tell us how your face checks are being attacked today. We will set up a sandbox and show you which checks fire and what comes back in the response.

Prefer to reach out directly?
Name *
Please enter a valid name
Company
E-mail *
Please enter a valid e-mail
Phone
Please enter a valid phone
Your Message (optional)
Please accept the privacy policy
FAQ

Deepfake Detection FAQ

Product questions. The background — how the attacks work and what regulators ask — is in our guide.

Didn’t find your answer?

AI-generated faces, swapped faces, fully generated images, deepfake video played back to the camera, video injected in place of the camera, and a second face in the frame. These are checks on the customer's face during verification; we do not analyze audio or media unrelated to the session.

No. The checks are steps of the KYC session, turned on in the scenario editor alongside the selfie and liveness steps.

A named check with its own result and confidence, plus an error code with a description from our public error reference.

No. Detection results depend on the attack, the device and the generator used, and a percentage quoted without those conditions is not something you could reproduce. We state what the checks look for instead.

Liveness decides whether a live person is in front of the camera. These checks decide whether the image of the face is genuine and whether the stream is. A deepfake shown to the camera is caught by both.

Face checks come with the Basic package, from $0.44 per check. Build your own set on the pricing page.

All-in-One Face Recognition Platform
for Instant KYC Verification

Streamline onboarding, prevent fraud, and ensure KYC/AML compliance