AI facial analysis that runs in the browser (no uploads, 17 metrics)(realsmile.online)

  The core idea: most face analysis tools upload your photos to a server.
  This one runs entirely client-side using face landmark detection — your
  photos never leave your device.

  It measures 17 structural facial metrics: canthal tilt (in degrees),
  facial width-to-height ratio, jawline angle, facial thirds balance,
  hunter eye index, golden ratio score, and more. All derived from 68
  facial landmarks detected locally in the browser.

  A few honest caveats HN will probably want to know:
  - The attractiveness correlations are from peer-reviewed research but
    the effect sizes are modest — this measures geometry, not beauty
  - Landmark detection accuracy drops on non-frontal photos or unusual
    lighting
  - The "ideal ranges" are population averages from Western-centric studies

  The looksmaxxing community (people trying to systematically improve
  their facial aesthetics) turned out to be the core audience — I didn't
  expect that going in.

  Tech: Next.js, face-api.js for landmark detection, runs on Vercel.
  The premium report generates a PDF client-side.

  Happy to answer questions about the technical approach or the landmark
  detection pipeline.