Check an image

POST /v1/check: a verdict, which detector fired, both detector scores and the face box.

POST https://api.checkobot.ai/v1/check

Request

Send the image one of three ways. The original bytes are scored as sent; don’t resize or re-encode first.

• multipart/form-data with a file field
• JSON {"image_url": "https://…"}: we fetch it (http/https, public addresses only)
• JSON {"image_b64": "…"}: standard base64 of the original file

Accepted: JPEG, PNG, WebP, AVIF, GIF (first frame), BMP and TIFF, up to 20 MB.

Images under 256 px on the shorter side, or JPEGs below about quality 50, are refused with a reason and aren’t metered.

To keep the image, add store=true. See Image storage.

curl

curl https://api.checkobot.ai/v1/check \
  -H "Authorization: Bearer $CHECKOBOT_API_KEY" \
  -F "file=@photo.jpg"

# or from a URL
curl https://api.checkobot.ai/v1/check \
  -H "Authorization: Bearer $CHECKOBOT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"image_url": "https://example.com/photo.jpg"}'

Response

200 response

{
  "id": "chk_Elp2uq1pt4qfqnDo",
  "is_ai": true,
  "fired": ["face_specialist"],
  "reason": "Face analysis found strong signs that the face was swapped or edited with AI.",
  "face_found": true,
  "content_credentials": { "found": false },
  "image": { "width": 512, "height": 512 },
  "model": "genai_face_v1_1",
  "bands_version": "0.5",
  "latency_ms": 612,
  "scores": { "full_image_specialist": 0.1912, "face_specialist": 0.974 },
  "face_box": [71, 100, 136, 136],
  "profile": "general",
  "image_stored": false
}

is_ai · boolean

The verdict: true is AI, false is “No AI detected”. false is not proof the image is untouched: the detectors miss some AI images.

fired · string[]

Which detectors crossed their threshold: full_image_specialist (whole image) and/or face_specialist (face analysis), plus content_credentials when the file carries signed Content Credentials that record AI.

reason · string

One line explaining the verdict, written for people.

content_credentials · object

The C2PA manifest embedded in the file, if any: { found, signer, verified, ai }. verified means the signer chains to the C2PA trust list and the image is unchanged since signing; only verified manifests can make the verdict ai. { found: false } is normal: most sites strip manifests, so it says nothing either way.

scores · object

Each detector’s raw score, 0 to 1. They’re on different scales, so don’t average or compare them; use is_ai and fired. face_specialist is null when no face of at least 64 px was found. Both are null when verified Content Credentials record AI: that decides the verdict, so the detectors aren’t run.

face_box · number[] | null

[x, y, w, h] of the face checked, in original pixels, or null.

model · string

The detector model. Store it and bands_version with the result; verdicts can change when either changes.

bands_version · string

The version of the thresholds used for this verdict.

image_stored · boolean

true when this image is being kept. See Image storage.

“No AI detected” isn’t proof an image is real: the detectors can miss some AI images.