Compare two face photos and return a similarity score from 0 to 100. Best for customer onboarding, duplicate-account checks, selfie-to-ID matching, and agent workflows that need a same-person decision. Input supports file path, HTTPS URL, data URL, or base64. Suggested business rule: score >=80 means high-confidence same person, but final thresholds should be calibrated by scenario.

Detect whether a face photo is a genuine capture or a spoof/AI-forged image. Returns risk level plus attack tag, covering screen replay, print attack, synthetic face, adversarial traces, multiple faces, and poor-quality captures. Best for lightweight anti-fraud before account opening, loan application, payment-risk step-up, or AI-agent human gate.

Detect whether a face video is genuine or a replay/deepfake/synthetic attack. Includes retry handling for transient upstream busy responses. Best for higher-risk eKYC, remote onboarding, transaction step-up, and Know Your Agent human-binding gates. Video limit: max 20MB; upstream recommends short clips <=20s.

Extract structured data from a Chinese national ID card image. Front side returns name, sex, ethnicity, birth date, ID number and address. Back side returns issuing authority and validity. Best for identity-data prefill, document digitization, and onboarding workflow automation.

Extract bank card number and expiry date from a bank card front image. Best for payment binding, card-information prefill, and bank-account verification workflows.

Extract structured fields from a Chinese driver's license main page: license number, name, vehicle class, address, issue date and validity dates. Best for auto insurance, car rental, fleet compliance, and vehicle-finance workflows. Supplementary page is not supported by this upstream endpoint.

Extract structured fields from a Chinese vehicle license. Main page returns plate number, VIN, owner, engine number and vehicle model. Supplementary page returns inspection and capacity-related fields. Best for auto loans, fleet onboarding, vehicle insurance, and collateral checks.

Analyze an image/video for risk and scene attributes. Supports portrait labels such as mask, hat, sunglasses, coercion, unconscious/asleep, phone use, headphones, tattoo, nudity, critical patient, facial mask; and environment labels such as multiple people, inside car, hotel room, car dealership. Best for compliance scene checks, loan/insurance evidence review, and anti-fraud triage. Async upstream job is submitted and polled automatically.