TrueFaceRequest API access →

For engineering teams

Deepfake detection API for identity and trust pipelines

Send a face image, receive per-face manipulation and AI-generated likelihood scores plus the forensic signals behind them. Built as a supporting signal inside KYC, HR, insurance and newsroom review — not as identity verification.

What the engine returns

Manipulation likelihood, per face

Filter and retouch traces: skin-texture variance, local geometry distortion around jaw and nose, and symmetry beyond the natural human range — measured for each detected face rather than for the image as a whole.

AI-generated likelihood, per face

Frequency-domain artifacts and GAN inversion error that diffusion models and GANs leave behind, reported as a probability with lower confidence flagged on heavily recompressed inputs.

Face-swap and deepfake indicators

Seam-level signals: colour shifts between face and neck, differing compression noise inside and outside the face region, and inconsistent illumination direction.

Confidence-gated reconstruction

For flagged faces the engine attempts a reconstruction and returns an approximation of the likely unedited face only when reconstruction confidence is adequate; otherwise it returns the detection report alone.

Integration shape

  1. 1. Request access. API access is contract-based. Submit an enterprise request with a work email from the business section of the homepage; onboarding returns the endpoint contract, authentication and volume terms.
  2. 2. Submit an image. Post the face image from your backend at the point where a profile photo or submitted portrait enters your workflow.
  3. 3. Consume per-face scores. Route on the score and signals — request an original or live capture, escalate to manual review, or continue — rather than auto-rejecting on a probability.
  4. 4. Or connect an agent. TrueFace also exposes an MCP server with OAuth-authorised tools (list_scans, get_scan, get_account_usage) for agent frameworks.

Illustrative response shape (Sample data — not a published contract)

{
  "scan_id": "SAMPLE_SCAN_ID",
  "faces": [
    {
      "face_index": 0,
      "manipulation_likelihood": "MOCK_VALUE",
      "ai_generated_likelihood": "MOCK_VALUE",
      "signals": { "skin_texture_variance": "MOCK_VALUE", "gan_inversion_error": "MOCK_VALUE" },
      "reconstruction": { "available": false, "reason": "confidence_below_threshold" }
    }
  ]
}

Field names and values above are placeholders for illustration. The exact schema, limits and SLAs are confirmed in writing during onboarding.

Scope and limits

TrueFace scores image integrity and synthetic generation. It does not confirm who someone is, match a face to a document, authenticate identity documents, perform liveness or presentation-attack detection, or screen sanctions and PEP lists. Heavy recompression, screenshots and social-media processing reduce confidence. Read the standards mapping in the KYC guide and the business FAQ.

API FAQ

  • What does the deepfake detection API return?

    Per detected face: a manipulation likelihood score, an AI-generated likelihood score, and the underlying forensic signals (frequency artifacts, GAN inversion error, face-swap seams, skin-texture variance, geometry and symmetry measures). Scores are probabilities, not binary verdicts.

  • Does it handle group photos?

    Yes. Detection and scoring run per face, so a group photo where only one person is filtered or synthetic is scored correctly rather than averaged into a single image-level verdict.

  • Can the API return an unfiltered version of a face?

    When a face shows signs of alteration, TrueFace attempts a forensic reconstruction and returns an approximation of the likely true appearance only when reconstruction confidence is adequate. Otherwise it returns the detection report alone.

  • Is this an identity verification API?

    No. It scores image integrity and synthetic generation. It does not confirm identity, match faces to documents, authenticate documents, perform liveness or presentation-attack detection, or screen sanctions and PEP lists.

  • Can AI agents call TrueFace directly?

    Yes. TrueFace exposes a Model Context Protocol (MCP) server with OAuth-authorised tools for listing scans, reading a scan and reading account usage, so agent frameworks can consume forensic results without a custom integration.

  • How do we get API access?

    API access is contract-based. Submit an enterprise access request from the business section of the homepage with a work email; the team returns the endpoint contract, authentication details, rate limits and volume terms during onboarding.