Identity checks rely on two things: a photo of an ID document and a selfie. Generative tools can now produce both. Compliance and onboarding teams need a way to flag uploads that deserve a manual look.
What fraudsters do
- Generated faces on otherwise real templates, so the portrait matches no real person.
- Edited fields — date of birth, expiry date or document number changed on a real card.
- Screens instead of documents — a photo of a document displayed on another screen.
What to check
Compression mismatches
A pasted portrait or retyped field compresses differently from the rest of the card. An error level analysis heatmap marks those areas. See what error level analysis is.
Capture history
A phone photo normally carries a camera model and capture time. Uploads that were saved by an editor, or have no capture record at all, get flagged for review rather than auto-cleared.
Printing and security features
Microprint, holograms and guilloché lines are hard for image models to reproduce. Blurry or smooth patterns in those areas are a warning sign.
Use forensics to triage, not to decide alone
No detector should be the only control in a KYC process. VerifyAI sorts uploads into clear, review and likely-tampered, and records a SHA-256 fingerprint and history for each check, so the decision is documented. Our KYC document verifier is free to try, and a developer API is available for workspaces.
For the strongest result, have applicants capture photos through a Verified Capture link, so the file arrives straight from the camera.