Built for identity, fraud & compliance teams

Your Identity Stack Has a Deepfake Blind Spot.

Q: How does deepfake detection work?

DeepDetector runs an ensemble of forensic classifiers that inspect facial and temporal inconsistencies, GAN and diffusion fingerprints, and audio artifacts imperceptible to the human eye and ear. Every verdict comes with a manipulation heatmap and confidence score: evidence your team can audit, not just a red or green light.

Q: What should enterprises look for in deepfake detection software?

Coverage of the generators in the wild today (not a stale benchmark), explainable and exportable results that stand up in compliance review, algorithm-level certification such as ISO/IEC 30107-3, deployment that fits your data-residency rules, and sub-second latency inside your existing KYC or moderation flow. Which is exactly how DeepDetector is built.

Q: Can deepfakes be detected in real time?

Yes. Waver analyses speech in real time and language-agnostically, and DeepDetector returns image and video verdicts in under a second, fast enough to sit inside a live onboarding or meeting-verification flow without adding friction.

Q: How is DuckDuckGoose different from a free deepfake detector?

Free tools typically run a single model trained on academic data. DuckDuckGoose runs a continuously retrained ensemble, attributes the generating model, and produces a forensic report that is auditable and defensible, built under EU law and deployable on-premise.

Q: Does deepfake detection meet EU regulatory requirements?

The platform is built GDPR-native, aligns with the EU AI Act's transparency and risk obligations, and supports NIS2 and eIDAS 2.0 workflows. Data can be kept entirely within the EU or deployed fully on-premise.

Synthetic identities and AI-generated IDs slip through KYC and liveness checks every day. DuckDuckGoose catches them in seconds, with court-ready evidence.

Get Started

Real-time deepfake detection scanning a biometric selfie

Face morphing attack detected by DuckDuckGoose deepfake detection

AI-generated face swap detected during identity verification

Deepfake detected — identity fraud prevented

Deepfake Detection chosen by Industry Leaders Around the World

Why global platforms trust our deepfake detection

Built to catch real attacks, and prove every flag.

Trained on the deepfakes
attackers use today

Trained on adversarial datasets from real fraud attempts, it catches manipulated selfies, fabricated documents, and liveness spoofs, cutting false negatives and manual review.

Live in your flows in hours

Drop-in SDKs and sub-second API responses that slot into your existing KYC, document, or liveness flow. No rearchitecting, no user friction.

≈

Detection accuracy on
real-world deepfakes

Competitor deepfake detection comparison

Alternative solution comparison — deepfake detection

Every flag comes with court-ready evidence

Every detection returns an explainable decision, a visual trace, and machine-readable output your fraud, legal, and compliance teams can act on and defend in review.

DeepDetector API — upload image for instant deepfake detection

Our deepfake detection products

Prove what’s real.
Flag what’s fake. In seconds.

Phocus

Review Deepfakes Without Writing a Line of Code.

Detect Deepfakes across Media Types

Waver audio deepfake detection — check your audio

Voice deepfake detection microphone — Waver

Waver

Detect deepfake speech instantly and accurately across 16+ languages

Detect Audio Deepfakes

DeepDetector highlighting manipulated facial regions in a deepfake

DeepDetector

Instant, on-premise deepfake detection for images and videos.

Detect Video & Image Deepfakes

Use cases

Deepfake detection for every identity and fraud workflow

From KYC onboarding to insurance claims and border control, DuckDuckGoose adds a deepfake detection layer to the identity verification workflows your teams already run.

Core use case

Model coverage

We detect the deepfake generators your attackers are using this quarter

Detection is only as current as its training data. DeepDetector and Waver are retrained continuously against the 40+ image, video, and voice models producing real fraud today, not a benchmark frozen in 2023.

VIDEOSora 2

VIDEOVeo 3

VIDEORunway Gen-4

VIDEOKling 2.0

IMAGEMidjourney

AVATARHeyGen

IMAGEFlux

IMAGEStable Diffusion

FACE SWAPDeepFaceLab

✓ detected✓ detected

+ 26 more model families · new generators added within weeks of release

Explainable by design

Every verdict comes with the evidence to defend it

Rivals hand your team a black-box score and a shrug. DuckDuckGoose shows its work: one liveness clip, broken into independent forensic layers your risk and compliance teams can read, verify and stand behind in an audit.

SRCLiveness video+

The exact clip your customer submitted. Every layer below is computed from this one capture. Nothing added, nothing outsourced to a third party.

01Face located+

We isolate and align the precise region we score (the red box). The verdict is about the face itself, not the background, pose or lighting.

02Phocus heatmap+

Our Phocus map shows where the model actually looked. The red and yellow pixels are the ones that drove the decision. You see the reasoning, not just a number.

03Noise residual+

We strip away everything the eye recognises and keep only the noise floor. A real camera sensor and an AI generator leave unmistakably different fingerprints here.

Manipulation probability

RAW

SRC · Liveness

01 Face

02 Phocus

03 Noise

The evidence behind the verdict3 independent signals · all agree

Temporal consistency

Blink timing and head motion stay physically inconsistent across frames. A genuine recording never does that.

Agrees with verdict

Latent embedding

The clip’s fingerprint lands inside our cluster of known AI-generated faces, nowhere near genuine captures.

Agrees with verdict

Frequency fingerprint

A frequency view exposes the regular pixel grid every generator leaves behind. Invisible in the image, unmistakable here.

Agrees with verdict

Benchmarked accuracy

Every accuracy number, with the benchmark attached

Benchmark 01

DFDC

Deepfake Detection Challenge

View benchmark

96.5% accuracy

Scored on the DFDC public benchmark, updated Q2 2026, 14.5 points above the public baseline.

Benchmark 02

FaceForensics++

HQ compression set

View benchmark

98.1% accuracy

High-quality compression set, Q2 2026, 16.1 points above the public baseline.

Benchmark 03

Celeb-DF v2

Cross-dataset eval

View benchmark

97.3% accuracy

Cross-dataset generalization test, Q2 2026, 15.3 points above baseline.

Benchmark 04

False positive rate

In-the-wild traffic

See methodology

0.01% FPR

Roughly one false flag per 10,000 legitimate checks, measured on live production traffic.

Built to Catch What Others Can’t

Why Teams Choose DuckDuckGoose for Deepfake Detection

Deepfake detection accuracy

Industry-Leading Accuracy: Image, Video & Voice

Phocus retains DeepDetector’s impressive 95% to 99% accuracy for detecting deepfakes in images and videos. Trust that our AI can catch even the most sophisticated fake content.

<5sec

Real-Time Deepfake Detection

Verdicts in Under a Second

Phocus provides instant or near-instantaneous results for images, videos, and speech, ensuring you can act quickly when deepfake threats emerge.

0.01%

False Positive Rate

Ultra-Low 0.01% False-Positive Rate

Phocus achieves an exceptionally low 0.01% false positive rate, ensuring you only receive accurate alerts. This means fewer disruptions and more confidence in every deepfake detection, minimizing the risk of genuine content being flagged mistakenly.

Integration patterns

One detection layer, three ways to wire it in

However your verification flow is built, DeepDetector slots in without a rebuild. Pick the pattern that fits your architecture.

See how it fits your stack

Media receivedincoming

PARALLEL · NO ADDED LATENCY

DuckDuckGoosedeepfake check

Liveness

Other checks

Decision engine

Accept / deny

See the cost it saves

Media received

Deepfake gatefakes rejected

Other checks

Decision engine

Accept / deny

Run a retrospective sweep

Retrospective sweepre-scan every onboarded account

3,240accounts

12fakes found

Clean accountDeepfake flagged

Re-scanning your existing user base surfaces synthetic identities that slipped through onboarding — 12 injected fakes caught this run.

What Our Partners Say

Trusted by the Ones Who
Have to Get It Right

Forensic Science & Technology
Public Safety and Justice

Netherlands Forensic Institute

"DuckDuckGoose’s explainable AI equips us to verify audio and video authenticity swiftly, ensuring manipulated media doesn’t impact legal cases. Their technology reinforces our commitment to accurate, reliable forensic evidence."

Zeno Geradts

Brazilian Bank & Fintech
Banking & Finance

Banco Daycoval

"Facing rising deepfake fraud, DuckDuckGoose’s state-of-the-art technology has fortified our verification process. We now protect our clients with confidence, staying ahead of sophisticated threats in an increasingly digital world."

João Oliveira

Sr. Applications Security Engineer

Identity Verification Solution Provider
KYC Selfie Analysis

DataChecker

"DuckDuckGoose has redefined our deepfake detection, providing us with explainable AI that fortifies our identity verification process. It’s not just about security—it’s about building unwavering client trust through transparency and innovation."

Michael Hagen

VP of Product

Dutch House of Representatives
Government & Public Services

Tweede Kamer

"As deepfakes become a greater threat to political discourse, DuckDuckGoose’s cutting-edge detection technology empowers us to act swiftly and decisively, ensuring the integrity of public information and safeguarding democratic trust."

Henk-Jan Eras

Advisor to the Dutch Second Chamber

In the press

The press that covers fraud, covers us

Our deepfake research and detection work is featured across the world’s leading finance, technology and security media, from the Financial Times to INTERPOL.

Questions buyers ask

Deepfake detection, answered

How does deepfake detection work?+

DeepDetector runs an ensemble of forensic classifiers that inspect facial and temporal inconsistencies, GAN and diffusion fingerprints, and audio artifacts imperceptible to the human eye and ear. Every verdict comes with a manipulation heatmap and confidence score: evidence your team can audit, not just a red or green light.

What should enterprises look for in deepfake detection software?+

Coverage of the generators in the wild today (not a stale benchmark), explainable and exportable results that stand up in compliance review, algorithm-level certification such as ISO/IEC 30107-3, deployment that fits your data-residency rules, and sub-second latency inside your existing KYC or moderation flow. Which is exactly how DeepDetector is built.

Can deepfakes be detected in real time?+

Yes. Waver analyses speech in real time and language-agnostically, and DeepDetector returns image and video verdicts in under a second, fast enough to sit inside a live onboarding or meeting-verification flow without adding friction.

How is DuckDuckGoose different from a free deepfake detector?+

Free tools typically run a single model trained on academic data. DuckDuckGoose runs a continuously retrained ensemble, attributes the generating model, and produces a forensic report that is auditable and defensible, built under EU law and deployable on-premise.

Does deepfake detection meet EU regulatory requirements?+

The platform is built GDPR-native, aligns with the EU AI Act's transparency and risk obligations, and supports NIS2 and eIDAS 2.0 workflows. Data can be kept entirely within the EU or deployed fully on-premise.

Get Detection Running in Your Stack This Week

Whether you're evaluating deepfake detection for the first time or looking to layer it into an existing stack, we can walk you through deployment options, integration timelines, and what detection looks like in your specific environment.

Get Started