Classification

Classification types run fraud checks on images. They return a verdict label and confidence score rather than extracted fields.

These types are used for identity fraud prevention, document fraud detection, and KYC liveness verification. For the full picture of what each check catches — and how to gate cases on the verdicts — see Deepfake Detection, Liveness, Synthetic & Tampered Documents, and Goal Rules.

---

deepfake_detection — Deepfake Detection

Analysis type: fraud check

Detects whether a face image or video frame has been synthetically generated or manipulated using deepfake techniques. Used to protect selfie / liveness flows from AI-generated faces.

Fields extracted

Validation rules

Example

const result = await client.verify.check({
  check: "deepfake_detection",
  file_url: "https://storage.example.com/selfie.jpg",
});
// result.output.label → "real"
// result.output.confidence → 0.97
// result.output.manipulation_regions → null

Reject the submission if label === "deepfake" or confidence < 0.80.

const { label, confidence } = result.output;
if (label === "deepfake" || confidence < 0.80) {
  throw new Error("Face image failed deepfake check");
}

---

document_synthetic_check — Synthetic Document Check

Analysis type: fraud check

Detects whether a document image (ID card, passport, bank letter, etc.) has been digitally fabricated, template-filled, or edited. Complements OCR extraction by flagging suspicious documents before field validation runs.

Fields extracted

Validation rules

Example

const result = await client.verify.check({
  check: "document_synthetic_check",
  file_url: "https://storage.example.com/submitted_aadhaar.jpg",
});
// result.output.label → "synthetic"
// result.output.confidence → 0.91
// result.output.risk_level → "high"

Run this check before kyc_aadhaar extraction to reject digitally fabricated IDs:

async function analyzeAadhaar(fileUrl: string) {
  // Step 1 — fraud gate
  const check = await client.verify.check({
    check: "document_synthetic_check",
    file_url: fileUrl,
  });

  if (check.output.label === "synthetic") {
    throw new Error(`Synthetic document detected (risk: ${check.output.risk_level})`);
  }

  // Step 2 — extraction
  return client.verify.document({
    doc_type: "kyc_aadhaar",
    file_url: fileUrl,
  });
}

---

face_liveness — Face Liveness Check

Analysis type: fraud check

Determines whether a face image was captured from a live person or is a spoof (printed photo, screen replay, 3D mask, or cut-out). Used in selfie + ID matching flows to prevent presentation attacks.

Fields extracted

Validation rules

Example

const result = await client.verify.check({
  check: "face_liveness",
  file_url: "https://storage.example.com/selfie_capture.jpg",
});
// result.output.label → "live"
// result.output.confidence → 0.99
// result.output.spoof_type → null

Spoof example:

// result.output.label → "spoof"
// result.output.confidence → 0.88
// result.output.spoof_type → "printed photo"

---

Combining classification checks

A typical KYC flow runs all three checks in sequence before extracting identity fields:

async function kycOnboarding(selfieUrl: string, idDocUrl: string) {
  // 1. Liveness
  const liveness = await client.verify.check({
    check: "face_liveness",
    file_url: selfieUrl,
  });
  if (liveness.output.label !== "live") {
    throw new Error("Liveness check failed");
  }

  // 2. Deepfake
  const deepfake = await client.verify.check({
    check: "deepfake_detection",
    file_url: selfieUrl,
  });
  if (deepfake.output.label === "deepfake") {
    throw new Error("Deepfake detected");
  }

  // 3. Synthetic document
  const synthetic = await client.verify.check({
    check: "document_synthetic_check",
    file_url: idDocUrl,
  });
  if (synthetic.output.label === "synthetic") {
    throw new Error(`Synthetic document (risk: ${synthetic.output.risk_level})`);
  }

  // 4. Extract identity fields
  return client.verify.document({
    doc_type: "kyc_aadhaar",
    file_url: idDocUrl,
  });
}