How FaceOracle Works

A step-by-step walkthrough of how your upload becomes an entertainment-only Style Mood Report.

FaceOracle is not a facial-recognition system. It is an entertainment service that interprets the visual vibe and mood of a single photo in plain language. This page walks through the full path, from upload to share link, so there are no surprises.

Step 1 — Consent before upload

A checkbox above the upload area takes explicit consent. You can upload only your own photo or one you have the subject's permission to use; under 14 requires a guardian's consent. The checkbox repeats that results are entertainment-only and must not be used for personality, ability, health, or identity judgments.

Step 2 — In-browser preprocessing

JPEG/PNG/WebP files are resized in the browser to a long-edge 800px copy, and a separate 200×200 thumbnail is prepared for optional share cards. Nothing leaves your device until preprocessing is complete.

Step 3 — Transport to our server

The resized image and your language code (ko/en) are sent to our Next.js route handler over HTTPS. The server does not write the original image to disk or any database; it is held only in process memory.

Step 4 — Analysis with Claude AI

The server forwards the resized image to the Anthropic Claude multimodal vision API. The model reads the image and returns text-only output in impression-based categories:

  • Image mood title, comment, and strength keywords
  • Playful reinterpretation of traditional face-reading ideas
  • Five first-impression keywords
  • Age vibe (not a medical age estimate)
  • Similar-vibe celebrities (mood matching against public celebrity names — not face recognition)
  • A four-season personal color palette
  • Hairstyle and makeup style suggestions based on face-shape impression

Claude does not build a numeric faceprint. The prompt explicitly forbids inferring sensitive attributes and deterministic fortunes.

Step 5 — Temporary storage for sharing

The result JSON and the 200×200 thumbnail are stored in Upstash Redis with a 24-hour TTL. They are deleted automatically once the TTL expires. Share links point at this temporary ID. You can request earlier deletion and we will remove it as soon as we can.

Step 6 — Display

An "entertainment-only" banner appears above and below every result page. Each result card spells out what it does not determine. For example, the similar-vibe celebrity card is explicitly labeled as not face recognition and not identity verification.

What we do NOT do

  • Build facial-recognition templates (faceprints). The output is natural-language interpretation, not a feature vector.
  • Infer sensitive attributes — gender, nationality, race, health, religion, or sexual orientation are disabled by default.
  • Produce decisions about hiring, admissions, lending, contracts, or security access. The service is not fit for those purposes.
  • Use face images for ad targeting. Images are not forwarded anywhere beyond the Anthropic API call.

Related reading

Back to Home·FAQ·Blog