PhotoCheck API

Face detection

See every face in a photo, where it is and how sure the detector is. Nobody is identified, and nothing is kept.

Drop a photo to check it

or paste one with ⌘V / Ctrl+V

JPEG, PNG, WebP or GIF up to 10 MB. Checked in memory, never stored.
Have an API key?

With a key, checks here spend your credits instead of the free daily allowance. It stays in this browser.

Detection, not recognition

PhotoCheck's face detection finds faces in an image and returns a box and a confidence for each one. It doesn't identify anyone, compare faces or guess age or gender. That's enough to count people, require a single face in a profile photo, or blur faces before publishing.

Why the difference matters

Common uses

Face detection API

POST /api/v1/faces takes a file, a base64 string or an image URL and returns the faces as JSON. Set min_confidence between 0.2 and 0.95 to tune it.

bash
curl https://photocheckapi.com/api/v1/faces \
  -H "Authorization: Bearer $PHOTOCHECK_API_KEY" \
  -F image=@group.jpg \
  -F min_confidence=0.4

Endpoint reference

200 OK (abridged)
{
  "faces": {
    "count": 2,
    "faces": [
      { "box": { "x": 412, "y": 188, "width": 166, "height": 166 }, "confidence": 0.93 },
      { "box": { "x": 120, "y": 240, "width": 131, "height": 131 }, "confidence": 0.81 }
    ],
    "min_confidence": 0.5,
    "model": "blazeface (Human 3.3)"
  }
}

Questions

What does face detection return?

For each face: a box in pixels (x, y, width, height) in the image's displayed orientation, and a confidence from 0 to 1. Faces are sorted most confident first.

Does it recognise people?

No. PhotoCheck detects faces only. It never computes an embedding, never compares faces, and never estimates age, gender or emotion.

What is min_confidence?

The cut-off below which faces are dropped. The default 0.5 keeps false positives rare; 0.3 catches more small and side-on faces but can mistake windows or patterns for faces.

Which model is it?

BlazeFace, run through the open-source Human library on our server. It's built for frontal and near-frontal faces and takes 10–30 ms an image.