Check an image

POSThttps://studioapi.kriyam.ai/models/defake/v2

Content type multipart/form-data

Tells you whether an image is AI-generated, manipulated or a real photograph, and checks any face in it for deepfake signs. The call is synchronous: you get back the finished run, already charged.

Request

curl -X POST "https://studioapi.kriyam.ai/models/defake/v2" \
  -H "X-API-Key: $KRIYAM_API_KEY" \
  -F "image=@photo.jpg"

Response

{7 items"id":"0c9d41a7-6b2e-4f10-93a5-77e2b0..."
"tool_url_path":"ai-image-detection"
"status":"completed"
"result":{1 item"aigc_analysis":{5 items"classification":"ai_generated"
"confidence_score":94.2
"watermark_detected":false
"watermark_details":{}0 items
"face_deepfake":{4 items"face_detected":true
"verdict":"fake"
"real_probability":0.08
"fake_probability":0.92
}
}
}
"error":null
"credits":2
"created_at":"2026-09-18T09:12:04.182741Z"
}

Authentication

Send your key in the X-API-Key header. See Authentication.

Request

Body

Name	Type	Required	Description
`image`	file	Required	The image to check. PNG or JPEG, up to 2 MB.

Response

200 OK

Returns the run envelope. The fields below are what this tool puts inside it.

Field	Type	Description
`result.aigc_analysis.classification`	string \| null	How the whole image was judged to have been produced.
`result.aigc_analysis.confidence_score`	number \| null	How sure that classification is, 0-100.
`result.aigc_analysis.watermark_detected`	boolean \| null	A generator's watermark was found — strong evidence, independent of the classifier's own judgement.
`result.aigc_analysis.watermark_details`	object	`{}` when nothing was found. On a detection it carries `confidence` — note that one is 0-1, unlike `confidence_score` — plus whatever else the detector reports. The watermark image itself is in `output_urls`.
`result.aigc_analysis.face_deepfake.face_detected`	boolean	Whether the face model had anything to judge. False means not applicable, not clean — every field below is null then.
`result.aigc_analysis.face_deepfake.verdict`	string \| null	`real` or `fake`, for faces within the image. This can disagree with `classification` — a face-swap on a real photograph fails here and passes there — so check both.
`result.aigc_analysis.face_deepfake.fake_probability`	number \| null	How likely the face is swapped or synthetic, 0-1.
`result.aigc_analysis.face_deepfake.real_probability`	number \| null	The converse, 0-1.

Errors

Status	Code	When
400	—	The image is empty, over 2 MB, or can't be decoded.

Every keyed call can also get401402403404429— see Errors.