Synthetic faces
Face swaps, morphs, and fully synthetic identities from a single image or prompt.
What is a deepfake
A face that never sat in front of a camera
A deepfake is a face generated or swapped by AI. The best ones fool people: humans catch them barely better than a coin flip. But the model leaves signals behind.
Synthetic · the signals give it away
- Blended edges Seams where a new face is stitched over the hairline and jaw.
- Lighting mismatch Highlights and shadows that don't match the real scene.
- Unnatural gaze Blinking, eye lines, and micro-expressions that don't behave.
- Over-smoothed skin Pore and texture detail the model can't fully reproduce.
Signals Deepsight reads in real time, the ones a reviewer can't.
The attack
How a fake face reaches your camera
Generative AI made fakes cheap and convincing: a believable one takes 35 seconds and a sub-$20 tool. Injection attacks skip the lens entirely, beating picture-only liveness.
Camera injection
Prerecorded or AI-generated video piped into the flow through a virtual camera.
Presentation attacks
Printed photos, screen replays, and 2D or 3D masks held up to a real camera.
0%
growth in fintech deepfake incidents in a single year
Deloitte, 2024
50-59%
human accuracy spotting deepfakes, barely better than a coin flip
Cooke et al., 2024How it works
Four trust layers. One score.
Deepsight checks every session across four trust layers at once, behavior, device, camera, and perception, then fuses them into a single score, in real time and invisible to the user.
Where it matters
Where deepfakes break digital trust
Fraudsters aim deepfakes at the moments that matter most. Deepsight runs invisibly inside each: a real customer sails through, a synthetic one is stopped.
-
Onboarding
Fraudsters submit deepfake selfies and injected video to open accounts at scale.
Real, live person -
Account recovery
Stolen IDs and prerecorded video slip past step-up checks to take over accounts.
Deepfake rejectedLive selfie VerifiedPrerecorded video BlockedStolen ID photo Blocked -
Helpdesk & agents
Manipulated faces and real-time swaps fool support agents on video calls.
Face swap detected -
Bot-driven farms
Scripted sessions on emulated devices flood verification, mimicking real users.
Bot farm flaggedFlaggedClearFlagged -
Workforce & hiring
Remote candidates use stand-in faces and deepfakes to pass hiring checks.
Deepfake stand-in blocked
Enrolled ID On video
Verified proof
Global banks, fintechs, and marketplaces trust Incode.
68×
better false-positive rate than the next-best commercial deepfake detector.
8 of 10
top U.S. banks choose Incode
4 of 5
top LATAM banks run on Incode
24
detection systems benchmarked by Purdue
What is deepfake detection?
Deepfake detection identifies AI-generated or manipulated media, synthetic faces, face swaps, morphs, and injected video, used to impersonate a real person. Incode analyzes biometric and visual signals in real time to confirm the person on camera is genuinely live and present.
How does Incode detect deepfakes?
Incode runs Deepsight, a multi-layer engine. It blocks device emulators and virtual cameras before capture, validates the camera source against injection, and uses multi-modal AI plus a vision-language model to read the artifacts generative AI leaves behind, all fused into a single score in milliseconds.
What is a video injection attack?
A video injection attack routes a prerecorded or AI-generated feed into a verification flow through a virtual camera driver, bypassing liveness checks that only analyze the picture. Incode detects injection at the signal level, not just the visual level, and blocks virtual cameras before capture begins.
What is liveness detection, and why isn't it enough on its own?
Liveness confirms a real, physically present person rather than a photo or replay. But modern deepfakes and injection attacks can fool liveness that only inspects the image. Deepsight adds device, camera, and behavioral integrity on top of perception, closing the gaps a liveness check alone leaves open.
How accurate is Incode's deepfake detection?
Benchmarked by Purdue University against the most widely used commercial detectors, Incode achieved the highest accuracy and the lowest false-acceptance rate. In production it delivers a 68× better false-positive rate than the next-best commercial technology, operating in milliseconds.
What's next
Stop deepfakes before they reach you.
See how Deepsight confirms a real, live person and blocks injection, spoofs, and synthetic media in real time.