AI photo check · $2.99
Is this photo of me AI‑generated?
Found a photo of yourself that you don’t remember taking? Upload it, and an image classifier we tested on 2,866 AI-generated images and 4,270 real photos checks whether it may be fully AI-generated, or generated from a reference photo of you.
- You see whether your photo can be checked before you pay.
- The answer comes with our tested numbers and their uncertainty.
- Your photo is not kept, and no faceprint is made.
Check a photo
The check runs first. If we can’t check your photo, there is nothing to pay. If we can, the result costs $2.99.
Never upload a sexual or intimate image of anyone under 18. If you found one, report it to NCMEC’s CyberTipline.
We don’t keep your photo. It is checked on our own servers and deleted when the check finishes; we keep the file’s SHA-256 fingerprint and the result.
What you get for $2.99
- One answer: “May be AI-generated” or “Not flagged”.
- The numbers behind it: how often the classifier caught AI-generated images and how often it wrongly flagged real photos in our tests, each with its 95% interval.
- A record you can share: the date, your file’s SHA-256 fingerprint, and the model and threshold used. We keep it for 90 days.
- Next steps if the photo may be fake: removal requests and where to report it.
What it cannot tell you
- Edited photos and face swaps. A real photo with a changed body, background or clothing, or your face placed onto someone else’s body, is not what this classifier detects.
- “Not flagged” never means real. In our tests it flagged 83.8% of fully AI-generated images, so it misses some.
- Who made it, or where it came from. The check looks at the image, nothing else.
- Explicit images and screenshots. Our numbers come from non-explicit photos of people. You can check an explicit image of an adult, but we have not measured how well the check does on one. Upload the original file rather than a screenshot.
- Every generator. It caught fewer images from Stable Diffusion 3 medium (76 of 211) and Midjourney v8.2 (114 of 169). Newer generators have not been measured yet: GPT-Image-2.5, Nano Banana 2 and Nano Banana Pro, FLUX.2, Seedream 5.0, Ideogram 4.0.
How we tested it
We wrote the test plan down before looking at any result, then scored images the model had never seen: 4,270 real photos of people from public collections and 2,866 images from 25 AI generators and setups, each saved as a JPEG at quality 90. Your photo is converted the same way, and the same model file, code and threshold we tested run your check. The classifier is the open haywoodsloan/ai-image-detector-deploy model.
Fully AI-generated images it flagged
83.8%
95% interval 67.1%-87.0% · 2,518 test images
Images generated from a reference photo it flagged
99.7%
95% interval 94.0%-100.0% · 348 test images
Real photos it wrongly flagged
30 in 10,000 (0.30%)
95% interval 18-52 in 10,000 · 4,270 real photos
Your photo
- Checked on our own servers and not sent to any other company. Payment is handled by Stripe, which never receives the photo.
- Deleted when the check finishes. We keep the file’s SHA-256 fingerprint and the result: unpaid results for 24 hours, paid results for 90 days.
- No face recognition. We detect where a face is, only to confirm the photo shows a person large enough for the check. We do not make a faceprint or compare the face with anyone.
If the photo may be fake
- Save the page address and a screenshot of where you found it before it changes.
- Ask the site to remove it. The TAKE IT DOWN Act requires covered platforms to remove intimate images shared without consent, including AI-generated ones, within 48 hours of a valid request.
- Find other copies of the image with a reverse image search, such as Google Lens, so you can report each one. FaceSentry’s free search looks for copies and similar images of a photo of you.
Questions people ask
Can this tell if a photo of me is a deepfake?
Partly. It can flag a photo that may be fully AI-generated, or generated by an AI tool from a reference photo of you. It cannot detect face swaps or edited real photos, which people also call deepfakes, so a result that is not flagged says nothing about those.
What does "Not flagged" mean?
That our classifier did not flag the photo. It is not evidence that the photo is real: in our tests it flagged 83.8% of fully AI-generated images, so it misses some, and it cannot see edits or face swaps.
Why does the check run before I pay?
So you never pay for a photo we cannot check. If the photo has no face large enough for the check, you see that first and pay nothing. If it can be checked, the result costs $2.99.
Can I show the result to someone else?
Yes. The result page records what was checked (the file's SHA-256 fingerprint), when, and with which model and threshold, and you can share its link or print it. It is an advisory signal from a classifier, not a determination; no classifier result proves how an image was made.
Do you keep my photo?
No. It is checked on our own servers and deleted when the check finishes. We keep the file's SHA-256 fingerprint and the result: unpaid results for 24 hours, paid results for 90 days.
Can I check an explicit image?
An explicit image of an adult, yes: it is handled the same way and never stored. Our tested rates come from non-explicit photos of people, so we cannot tell you how well the check does on explicit images. Never upload a sexual or intimate image of anyone under 18: report it to NCMEC's CyberTipline (report.cybertip.org). If you are under 18, NCMEC's Take It Down service (takeitdown.ncmec.org) can help remove images of you.
What if I pay twice or the check expires?
A second payment for the same check, or a payment for a check that has already expired, is refunded in full automatically. For anything else, email support@facesentry.com.
An advisory signal from an image classifier, not a determination. General information, not legal advice.