Generated face evidence
Deepfakes can create a convincing portrait or alter the identity shown in a video. FaceX’s deepfake models assess face evidence for synthetic or manipulated content.
SYNTHETIC MEDIA DEFENSE
A convincing face.
A closer examination.
FaceX combines deepfake detection models trained from scratch with camera authenticity checks, anti-spoofing and a risk engine. Captured face evidence is assessed for manipulation before it supports an identity decision.
Models trained from scratch
FaceX-developed detection models.
Manipulation assessment
Evidence of generated or altered media.
Connected decisions
Detection signals feed the risk engine.
THE PROTECTION LAYER
Deepfake detection, explained.
Identity substitution
A face that resembles the reference can still be artificial. Deepfake detection adds an authenticity assessment alongside the identity comparison.
Layered verification
The media assessment is considered with camera integrity and anti-spoofing signals. One reassuring result should not cancel a critical integrity failure.
FROM CAPTURE TO CONFIDENCE
How the evidence connects.
Follow face evidence into the detection layer and then into a combined decision. The diagram explains the security relationship; it does not reveal model internals or decision thresholds.
THE EVIDENCE JOURNEYConceptual security flow
Captured faceImage or video evidence
Live sessionCapture context
Identity referenceComparison evidence
Deepfake assessmentManipulation evidence joins the security checks
Combined risk decisionRequired checks and unresolved risks considered together
ApproveRequired evidence is consistent
ReviewMore evidence or attention needed
RejectBlocking risk or failed requirement
SEE THE DIFFERENCE
Evidence changes the outcome.
Choose a scenario to see how different signals can affect an example decision.
Illustrative scenarios only. These qualitative signals and policies do not expose production scores or thresholds. No camera access, biometric capture or live check occurs on this page.
PERFORMANCE CONTEXT
A reported result.
A clearly defined claim.
99%
FaceX-reported deepfake detection figurePerformance needs a test context.
The 99% figure is reported by FaceX. Public benchmark methodology and independent validation are not available on this page. It is not a guarantee that every deepfake, injection or presentation attack will be blocked.
Read the metric with its conditions
A useful benchmark identifies the model version, test size, attack types, devices, capture conditions and decision threshold. Detection rates can change when those conditions change.
Measure both kinds of error
Report missed attacks alongside genuine-user rejection rates. A detector’s performance and the complete verification system’s performance are separate measurements.
COMMON QUESTIONS
A clearer understanding.
Does a strong face match prove the media is genuine?
No. Matching assesses resemblance to an identity reference. Manipulation detection, camera authenticity and anti-spoofing address separate risks.
Does 99% mean every attack is blocked?
No. The 99% figure is FaceX-reported performance, not a universal guarantee. Its meaning depends on the evaluated attacks, test conditions, decision threshold and errors on genuine users.
What happens when evidence is inconclusive?
The risk engine can route an inconclusive result for review or further capture according to the integration’s decision policy. The scenarios here illustrate that distinction.
Further reading: NIST guidance on digital injection and forged media and NIST presentation attack detection evaluation.These sources explain the threat categories and evaluation concepts. They do not certify FaceX or validate its reported performance.