Fake Image Detector
Take another look at a suspicious image — start with the pixels.
JPEG, PNG, WebP or GIF (first frame) · up to 5 MB · Imagefake copies expire within 24 hours
1 free check. No account needed. Then $2.99 per month for 20 checks — see pricing.
Three kinds of "fake"
Fully AI-generated. The depicted scene was synthesized. Inconsistent text, anatomy or background detail may deserve review, but visual clues alone cannot establish how it was made.
AI-edited. A real photograph may have a region replaced, an object added, a background extended or a face swapped. Most of the picture may remain photographic, so an overall AI / human prediction may miss the edit.
Old-school manipulation. Cloning, pasted objects, changed document values and spliced screenshots do not require generative AI. This classifier does not certify the absence of these changes.
What this detector checks for
AI-generation classification. Imagefake submits the prepared still image for an AI / human prediction. Its confidence reflects the model output, not a verified probability that a photo is genuine or doctored.
Noise, compression and clone traces: manual review. Compare texture, seams and repeated patches yourself or with a qualified examiner. Imagefake does not currently calculate a forensic noise map or clone/splice report.
Documents, receipts and screenshots: compare the source. Look for inconsistent fonts, alignment or content, and verify with the issuing party. A classifier cannot validate a transaction or document claim.
Faces and content credentials: current limits. This service has no specialized face pipeline or C2PA trust-list verifier. A still image cannot show blinking or motion. Use original source material and a credential verification tool when provenance matters.
What the report shows
The prediction. AI-generated or human-made, with model confidence. It does not return a separate edited/upscaled verdict or prove that no manipulation occurred.
The check record. An image preview, check time and available classification output. Export JSON or print before the report expires. There is no region-by-region heatmap, metadata history or C2PA status block.
Further review. Compare original files and earlier versions, verify the source and consider the claim made about the image. For artwork, the AI art detector page provides guidance on process evidence.
When you'd actually use this
Buying or selling second-hand. A suspicious listing or damage photo can prompt a source check. Ask for additional original photos and use the platform’s dispute process; do not treat the classification as verification of the item.
Expense claims, receipts and documents. Compare versions and verify amounts with the source. The service is not a financial or document authentication tool.
A photo that arrived out of context. Look for the original publication, date and location before accepting the caption. A real photograph can carry a false claim.
A face you do not recognize. A still-image score cannot establish identity or trustworthiness. Avoid uploading sensitive images, and use independent verification before sending money or personal information.
Frequently asked questions
Can this detect a photo that was AI-edited, not fully generated?
It may produce an AI prediction, but it does not reliably identify localized edits or distinguish them from full generation. A human-made prediction does not rule out alteration.
Do deepfake stills get detected?
Some may produce an AI prediction, but this is not a specialized deepfake or identity detector. There is no face-motion or blink analysis, and video is unsupported.
What about a screenshot of a photo?
A screenshot changes the file and may remove useful metadata or detail. Results need caution. If you can obtain the original file, review that and look for earlier versions.
I know a photo was altered but the result says human-made. What now?
Do not treat the prediction as proof. Compare the altered and original versions, verify with the source and seek specialist review when appropriate. A report is only a dated model output, not legal evidence; read terms.
How is this different from the homepage checker?
It uses the same classification as the homepage checker. This page adds guidance about edited and misleading images, not additional forensic detection. All three tools share the same free allowance.
How many free suspicious-image checks do I get?
1 free check. No account needed. Then $2.99 per calendar month for 20 checks. Active subscribers can buy 10 permanent credits for $1.49; those credits remain usable after the subscription ends. Checks are subject to available capacity. Our privacy policy explains the separate Imagefake and provider retention policies.