KYC and onboarding
KYC AI detection: check ID photos and selfies for AI-generation evidence
isGenAI screens the ID document photos and selfies customers upload during KYC for evidence that AI made or edited them: C2PA Content Credentials from AI generators, IPTC AI labels, saved generation prompts and generator metadata. Free, through a web checker or an API with no key.
What it checks in a KYC upload
- C2PA Content Credentials. Images from generators that sign their output record the tool and an AI source type. isGenAI reports signature validity and signer trust separately.
- IPTC digital source type. Labels such as
trainedAlgorithmicMediaorcompositeWithTrainedAlgorithmicMediawritten by AI tools and editors. - Saved prompts and settings. Prompts, seeds and model names that AUTOMATIC1111, ComfyUI, InvokeAI and NovelAI save inside the file.
- Generator and editing metadata. Software and history fields that name an AI tool, and missing camera fields where a phone photo would normally have them.
- Known copies. Matches with images already checked in isGenAI records.
Every finding names the field it came from, so a reviewer can see why a file was flagged.
Where it fits in onboarding
Run the check when a customer uploads an ID photo or selfie, before or alongside your identity verification provider. Route files with AI-generation evidence to manual review. Keep document security checks, liveness and face matching with your KYC vendor: isGenAI does not perform them.
curl --fail-with-body --max-time 30 https://isgenai.com/v1/check \
-F "file=@id-front.jpg" \
-F "privacy=private" -F "public_receipt=false"The same checks run from the MCP server for AI agents, and from the web checker for one-off reviews. For video selfies, the video checker reads Content Credentials and container metadata. Fraud checks beyond onboarding, such as receipts and claim photos, are in the AI fraud detection guide.
Limits that matter for fraud teams
Screenshots, re-exports and messaging apps remove most of these records, and determined fraudsters strip them on purpose. A clean result therefore says nothing about whether a document or face is real. isGenAI publishes no accuracy percentage and gives no visual verdict on a forged document.
KYC AI detection questions
Can isGenAI detect an AI-generated ID document?
It finds AI-generation evidence stored in the uploaded file: a C2PA manifest from an AI generator, an IPTC digital source type such as trainedAlgorithmicMedia, prompts and settings saved by Stable Diffusion tools, and generator metadata. When a file carries one of these, the result names the record. It does not judge whether a document looks forged from its pixels.
Does a clean result mean the ID or selfie is genuine?
No. Fraudsters often screenshot or re-export images, which removes these records, and a missing record is not proof of a camera or a real person. Treat an isGenAI finding as a reason to escalate, and keep document authenticity, liveness and database checks with your KYC provider.
How do I add it to an onboarding flow?
POST each uploaded image to https://isgenai.com/v1/check with privacy=private and public_receipt=false. No API key or account is needed. The JSON response lists every finding with its source, so your rules engine can route files with AI evidence to manual review. Capacity is shared, so respect HTTP 429 and Retry-After.
Are ID images stored or published?
Private uploads are processed on isGenAI’s server and stay out of public reports and receipts. Send privacy=private and public_receipt=false for every identity document, and follow your own data-protection obligations before sending personal data to any third-party service.
Does isGenAI replace a KYC or identity verification provider?
No. isGenAI does not verify identity, run liveness or face-match checks, or validate document security features. It adds one signal those systems often skip: whether the file itself records that AI made or edited it.