Image Background Remover

Cuts the person out of a photograph and returns it without the background, for ID photos and profile pictures.

POST/v3/store/smart-city/remove-backgroundTry it

• Price1 ICper request
• Measured0.790
• Served680.7Kcalls

• v2.0
• Active
• Benchmark paper (PDF)

Sign in to run this demo.

50 credits free on sign-up. No card.

Create a free accountLog in

Request

curl -X POST https://api.iapp.co.th/v3/store/smart-city/remove-background \
  -H "apikey: YOUR_API_KEY" \
  -F "file=@selfie.png"

Response

401No API key in the request, or the key is not valid.

The body is the gateway's own and is not reproduced here.

Request

multipart/form-data, one image per request. 1 IC per request.

Headers

apikeystringrequired

Your API key, created under API Keys in the control panel.

Body

filefilerequired

The photograph. JPEG or PNG, up to 2 MB.

rotateIfPortiatboolean

Meant to rotate a portrait image to landscape in the output, but it currently has no effect on file uploads: the image comes back in its original orientation. The field is spelled as the API spells it.

The legacy paths /face-extractor and /remove-bg answer at the same price.

Response

The photograph with its background removed.

The photograph with its background removed.

Details

Measured

On the production service, 28 October 2026, against 802 photographs from the public COCO 2017 validation set, each showing one clear subject outlined by human annotators: 186 of people, 143 of animals, 218 of vehicles and 255 of other objects. The score is IoU, the overlap between the cut-out and the true subject, where 1 is perfect.

Measure	Result
Mean IoU, 186 photographs of people	0.790 (95% CI 0.760–0.818); median 0.857
People filling 10–20% / 20–30% / 30% or more of the frame	0.815 / 0.822 / 0.743
Failed cut-outs of people (IoU below 0.5)	20 of 186, 15 of them people filling 30% or more
Mean IoU, all 802 photographs	0.733 (95% CI 0.714–0.752); median 0.856
Animals / vehicles / other objects	0.825 / 0.722 / 0.651
Failed cut-outs, all 802 photographs	149 (18.6%)
Median time through the API, 150 photographs	0.33 s

Most cut-outs are close, and the failures are total: in a dark or low-contrast scene the service can keep almost nothing, dark clothing against a dark background can be lost while the face is kept, and a person holding an object such as a cake or a telephone close to the camera can be dropped in favour of the object. A close-up scores lower than a person at mid distance, so leave space around the person rather than filling the frame. The test photographs are consumer photographs of at most 640 pixels with hand-drawn outlines, not studio portraits on a plain background, so the ID-photo case and fine edges such as hair are not measured on their own. Method and limits: the benchmark (PDF).

Limits

• Made for photographs with one clear subject, such as a person. A close-up is cut out less well than a person at mid distance; see Measured.

Data handling

Images are processed in memory and not kept after the response. Consent and notice under PDPA and GDPR remain with you as the data controller.