What is a reverse image search?
A reverse image search starts with a picture and asks where else that picture appears. It is the opposite of an ordinary image search, which starts with words and looks for pictures that match them.
The difference decides what the tool can answer. Given a profile photo, an engine can tell you the same picture was published on a modelling portfolio in 2019 under another name. It cannot tell you who is in the picture: TinEye states in its own help pages that it does not recognise faces in the photos, and that is normal behaviour rather than a shortcoming.
That is the right shape for checking a stranger. You are not trying to identify a face, which is a harder and more invasive question. You are asking whether the file in front of you has a history the person showing it to you did not mention.
Reading what the camera wrote inside the file is a separate job, and the guide to photo metadata covers it.
How does a reverse image search work?
By fingerprinting the pixels rather than reading the file. TinEye documents the step plainly: it creates a compact digital signature for the image you submit, then compares that signature against every image in its index.
Two consequences follow, and both surprise people. The file name does not matter, and neither does the metadata, because TinEye does not use image names or any metadata attached to the picture you submit. Renaming a photo before you search it changes nothing at all.
The second is the ceiling on the whole technique. An engine can only return pictures it has already crawled and fingerprinted, so a reverse image search is a search of an index rather than a search of the internet. Anything behind a login, inside a private album, or never published in the first place sits outside it.
How do you check a profile photo, step by step?
Start with the largest copy of the picture you can get, because detail is what the fingerprint is built from.
- Save the photo at full size. Open the image in its own tab first: the thumbnail in a feed is a smaller and more heavily compressed file than the one behind it.
- Upload it. Google accepts a file upload, a drag-and-drop onto the search box, or a pasted image URL.
- Open the result pages, not the thumbnails. A visually similar picture is not the same picture, and the two are listed together.
- Run it again elsewhere. One engine is one index.
A screenshot works, but crop the interface out of it first. The app furniture around a profile picture is part of the image an engine fingerprints, and it is the part that no published copy shares.
A generated face has no earlier copy anywhere, which is why an empty result settles nothing.
Why does a reverse image search find nothing?
Usually because the copy you are holding is not the copy that was indexed. A fingerprint is built from the pixels, and three ordinary things change the pixels enough to break a match that genuinely exists.
- A crop. A profile picture is often a crop of a larger photograph, and the larger photograph is the one that was published.
- A mirror. Some apps and some front cameras flip an image horizontally without being asked. To an engine comparing pixel patterns, a mirrored face is a different face.
- Recompression. Every platform re-encodes what you upload, so a picture that has passed through three platforms is three generations away from the file that was crawled.
The other reason is coverage. Engines crawl different corners of the web and hold different indexes, so nothing found on one engine is not nothing found. An empty result is a statement about one index on one day.
Crop to the face and run it again
This is the single fix that rescues most failed searches, and it takes about ten seconds. Google Lens has a region box: drag the corners around the part of the picture you want, and the search runs on that region instead of the whole frame.
It works because the background is doing most of the talking. A photograph of a person in a room is mostly room, and the room is what the fingerprint describes. Crop to the face and shoulders and you throw away the part that differs between the published copy and the one you were sent.
Two further passes are worth the effort when the first fails. Flip the image horizontally in any photo editor and search the mirrored version. Then find the largest copy of the picture you can and search that, because a re-uploaded thumbnail has lost detail the original still carries.
Does reverse image search work on an AI-generated photo?
No, and the reason is structural rather than a gap somebody will close later. A face generated by a diffusion model has never existed as a published file, so no crawler has ever fingerprinted it and no index contains a copy to match.
That makes a clean result actively misleading. A generated face and a genuine private snapshot both return nothing, because neither was ever indexed. A reverse image search that finds nothing is not evidence that the person is real.
What helps instead is provenance rather than matching. Google's About this image panel surfaces C2PA content credentials and SynthID watermarks where a generator wrote them. Both depend on the tool that made the picture choosing to mark it, and marks can be stripped, so their absence proves nothing either.
A clean result is not a clearance. Treat nothing found as the search having no opinion. It moves you on to the other identifiers rather than settling anything.
Where the engines disagree
There is no best engine, only different indexes. Each row below is what the vendor documents about its own product, because a claim a company makes about itself is the only one it has to stand behind.
| Engine | What its own documentation says | The limit that follows |
|---|---|---|
| Google Lens | Search by file upload, drag-and-drop or image URL, with a region box for part of a picture | Similar pictures sit beside true matches; open the page, not the thumbnail |
| TinEye | Fingerprints the pixels and finds copies that were cropped, edited or resized | No face recognition, so another photo of the same person is a different image |
| Yandex Images | Returns exact copies of your image and similar images | Similar is not the same; treat each hit as a lead |
| PimEyes | A face search engine rather than an image matcher, with an opt-out | Searches for a face rather than a file, a different and more invasive question |
Run at least two of them. The cost is a minute, and the failure mode of running one is a false all-clear.
What does a match actually prove?
That the picture is older than the account showing it, and little more. That is still a great deal when the account presents the picture as its own.
Read the page rather than the result. The useful facts sit on the page holding the older copy: when it was published, what name is attached, what language it is written in, and whether the site is a stock library, a modelling portfolio or somebody's dormant profile from 2014.
Three innocent explanations account for most matches, and they are worth ruling out before you accuse anybody. The person has other accounts, which is ordinary. The picture is a stock photo used knowingly, or a scraper site copied a real profile belonging to somebody who did nothing at all.
A match, plus a young account, plus a story that needs money is a different picture from a match alone. Verifying someone you met online sets out the order to check the rest in.
The legal line, and what DetectiveCheck does not do
Uploading a picture somebody sent you to a search engine is lawful in the United States, the EU and the UK. The regulated part is what happens next: a photograph of an identifiable person is personal data under the GDPR, so publishing a match or contacting the people on the page you found needs more thought than the search did.
Face search is a different category from image matching, and worth treating as one. An engine that searches for a face returns pictures of a person who never posted the one you uploaded, which is why PimEyes publishes an opt-out at all.
Our own photo module is not a reverse image search, and the photo analysis page says so before you upload anything. It reads what is written inside a file and performs no face recognition, deliberately. For finding where a picture appears, use Lens or TinEye, which are free and better at it than we would be.
Common questions
Does a reverse image search work on a social media profile photo?
Often, but not always. Platforms resize and re-encode what you upload, so the copy on a profile is not identical to the one published elsewhere and the fingerprint an engine builds from it differs. Save the picture at the largest size the platform will give you, crop to the face, and run it on more than one engine before you conclude anything.
Why does a reverse image search find nothing?
Two reasons. Either the copy you searched has been cropped, mirrored or recompressed enough that its fingerprint no longer matches the indexed original, or the picture was never indexed at all: it sits behind a login, was never published, or was generated rather than photographed. Nothing found describes one index on one day, not the whole internet.
Does mirroring or flipping a photo defeat a reverse image search?
It can, because the engines compare pixel patterns and a mirrored image is a different pattern. Some apps and front cameras flip pictures without being asked, so this happens by accident as often as on purpose. The fix costs nothing: flip the image back horizontally in any photo editor and run the search again.
Does reverse image search work on AI-generated photos?
No. A face produced by a diffusion model has never been published, so no crawler has fingerprinted it and no index holds a copy to match. The search returns nothing, which is exactly what it returns for a genuine private photograph. A clean result is therefore not evidence that the person on the other end exists.
Which is better, Google Lens, TinEye or Yandex?
They answer slightly different questions, so run more than one. Google Lens has the broadest web coverage and lets you search a region of an image. TinEye fingerprints the pixels and is built for cropped and edited copies. Yandex returns exact copies alongside similar images. Disagreement between them is normal, and is why one result is not a verdict.
Is a reverse image search legal?
Searching with a picture somebody sent you is lawful in the United States, the EU and the UK. The restrictions are on use. A photograph of an identifiable person is personal data under the GDPR, so publishing what you found, or contacting people on the pages the search returned, carries obligations that the search itself does not.
Does the person know I reverse image searched their photo?
No. The search runs against an engine's own index of pages it has already crawled, and it does not contact the person, the platform they posted on, or any site in the results. Nobody is notified and nothing appears in anyone's account activity. Opening a result page is an ordinary visit to that page.
Can I reverse image search a screenshot?
Yes, and it often works, but crop it first. The interface around the picture, meaning the status bar, the buttons and the name above it, is part of the image an engine fingerprints, and it is the part no published copy shares. Crop tight to the photograph itself, then tighter to the face if the first pass fails.
How many variations should I try before giving up?
Four passes cover most of it: the picture as you have it, cropped to the face, mirrored, and the largest copy you can find. Run each on two engines. If all of that returns nothing, stop searching the picture and check the person's handle and email address instead, because those reach a great deal further.
