Face Swap Check: Tell If a Photo’s Face Was Swapped

0

๐Ÿ”Ž Reverse Image Search

Image search techniques made simple: upload a photo or paste a URL and search 6 engines at once. 100% free, nothing stored.

๐Ÿ”’ Your image is never stored on our servers longer than 24 hours and is never shared.

Yes, a face swap check photo test works, and you do it in stages. Zoom into the face boundary, compare the light on the face with the light in the room, check the ears and hairline, then reverse search for the original. No single check settles it. A photo passes when the pixels, the provenance and the source agree.

A face swap is not the same as a fully AI-generated photo. In a swap, software lifts one person’s face and grafts it onto another person’s head inside a real photograph. The room, the body, the clothes and the camera are all genuine. Only the face is borrowed. That is why the usual AI detectors struggle with swaps, and why the checks in this guide look at the join, not the whole picture.

Face swap, defined: a photo edit in which the face of one person is replaced with the face of another person, while the rest of the photo stays as it was taken.

This face swap check photo guide covers still photos only. Swaps in video behave differently, and our guide to checking a viral video clip covers that side. The method below works on a dating profile photo, a listing photo, a screenshot someone sent you, or a photo in the news. It does not teach you how to make a swap. It teaches you how to catch one.

What is a face swap, and how is it different from a fully AI-generated photo?

A fully AI-generated photo is invented from scratch, pixel by pixel. A face swap is surgery on a real photo. The distinction matters because it decides which checks work. A detector trained to spot fully generated images looks for generation signals spread across the whole frame. In a swap, most of the frame is a genuine photograph, so those signals are weak or absent. The swap leaves its traces in one place only: the boundary where the new face meets the old head.

Think of it this way. A fully generated photo has to invent a world that agrees with itself. A swap has to agree with someone else’s world: light the forger never controlled, a camera angle they did not choose, skin the model never saw. The checks below exploit exactly that mismatch.

Swaps turn up in ordinary places. A dating profile wears a stolen face. A marketplace listing shows a seller who does not exist. A political photo puts a candidate somewhere he never was. The photo in our worked example below did exactly that in January 2023, and it was shared hundreds of thousands of times before fact-checkers caught it.

Why are swapped faces so hard to spot?

Because the parts your eye trusts most are the parts that stay real. The background is real, the body language is real, and the photo has the grain and noise of a genuine camera file. Your brain reads the scene as a photograph, then accepts the face that sits inside it. A good swap survives a casual glance every time. Most swaps in circulation are not well made, though, and even the strong ones keep failing in the same predictable places.

There are three reasons swaps are hard. First, the face is small. On a phone screen, the join between the face and the head is a few millimetres wide, and your thumb covers it. Second, compression hides the evidence. Every re-upload and screenshot smooths the seam a little more. A detector test in 2023 by Bellingcat found that a leading image detector missed 7 of 10 AI images after social-media-level compression. Third, whole-image detectors are the wrong tool. They answer the question “was this whole image generated?” and a swap is not a whole generated image. Our worked example proves the point: a real, documented face swap scored 0% likely AI-generated on the Hive demo.

So the honest starting point is this. A clean result from one tool proves nothing. You check the photo in layers, and you only trust a photo when several independent checks agree.

What is the face swap check photo method?

The face swap check photo method is a six-part inspection you run in order, from the fastest visual tells to the checks that need a tool. Run them all. Each one takes seconds, and together they catch most casual fakes. The table below is the whole method in one place. The sections after it explain each row.

CheckWhere to lookWhat gives a swap away
1. The boundaryHairline, jawline, neck, earsA sharpness or texture change where the new face meets the old head
2. The lightNose shadow, chin shadow, catchlights in the eyesFace lit from a different direction than the room
3. The earsBoth ears, ear shape against a known photoEars that do not match the person, or one ear sharper than the other
4. The skinCheeks, forehead, neck below the jawSkin that changes texture or colour at the cheek line
5. The proportionsHead against shoulders, neck width, glassesA head that is too big or too small for the body wearing it
6. The originalGoogle Lens and TinEyeThe same photo online with a different face in it

1. Zoom into the boundary

Open the photo full size and zoom to 100% on the face. Look at the hairline first, then the jaw, then where the face meets the neck. A swap is pasted in, so the edge of the pasted region is where it fails. You are looking for a faint halo, a band of lighter pixels around the head or neck, a sudden change in sharpness, or skin that looks airbrushed on one side of a line and grainy on the other. Real skin has pores and the photo has grain. A grafted face often carries subtly different noise than the neck below it.

Check the glasses too, if the person wears them. Frames that cross the face are hard for swap software to reproduce, so they smear, clip or vanish where they meet the new face. A frame that stops dead at the cheek, or a lens that bends oddly, is a strong tell. A halo edge around a pasted face is one of the classic signs of an edit, and our guide to telling whether a photo is Photoshopped lists the rest of them.

2. Compare the light on the face with the light in the room

Find the light source in the scene. Look at the shadows on the wall, the table or other people, then look at the shadows on the face. If the room is lit from the left but the nose casts its shadow to the left as well, the face was lit somewhere else. Check the catchlights, the small bright reflections in the eyes. In a real photo, both eyes usually catch the same light from the same direction. If one eye has a bright window reflection and the other has none, or the reflections point in different directions, the face came from another photo.

3. Check the ears and the hairline

Swap tools replace the face, not the ears. The ears in the photo usually still belong to the original person, so compare them with a known photo of the person the face claims to be. Ears differ a lot between people, and they are hard to fake because the model pays them little attention. Look at the hairline as well. Hair is difficult to blend, so a swapped face often has a hairline that does not sit naturally on the head, or hair that changes texture right where it meets the forehead.

4. Read the skin across the cheek line

Skin is the giveaway that survives even a good swap. Run your eye from the cheek down across the jaw to the neck. The face and the neck should share the same tone, the same texture and the same grain. If the cheeks look smooth and poreless while the neck shows stubble, moles or texture, the face was probably smoothed by the swap and the neck was not. A colour step at the jaw, where the face is a shade warmer or cooler than the neck, means the two parts were photographed under different light.

5. Measure the proportions

Step back and look at the whole figure. Does the head fit the body? A swapped head is often slightly too large or too small for the shoulders under it, because the source face and the target body were photographed at different distances. The neck is a useful ruler: if the neck looks too narrow for the head, or the head sits at an angle the neck cannot explain, something was pasted. In our worked example, the fact-checkers flagged exactly this: the head was out of proportion to the rest of the torso.

6. Reverse search for the original

The strongest check is not visual at all. A swap is built on a real photo, and that real photo is often still online with the original face in it. Crop the photo tightly around the person, then run it through Google Lens and TinEye. If the same scene appears with a different face, you have your answer. The next section explains the steps, and the worked example shows what the results actually look like.

How do I reverse search a photo to find the original?

Reverse search is the check that turns a suspicion into proof, because it finds the photo the swap was built from. Follow these steps in order.

  1. Save the best copy you can. Ask the sender for the original file if you can. A screenshot of a screenshot loses the detail a reverse search needs.
  2. Crop tightly around the person, but keep the background in the frame. The background is what matches. A face-only crop matches every face that looks similar, which is useless.
  3. Run the crop through Google Lens. Open the exact matches tab and read the earliest results. Look for the same scene with a different face, or the same photo on a fact-check site.
  4. Run the same crop through TinEye. TinEye reports how many matches it found and the earliest date it indexed each one, which tells you which copy came first.
  5. Compare dates. The original is usually the earliest dated copy. A photo that first appears on a fact-check page with a fake stamp on it is not the original. It is the debunk.
  6. Do not stop at a zero. A photo that is new, cropped or flipped may have no matches yet. A zero on TinEye or Lens never clears a photo. It only means the search found nothing.

If you find the original, save the link and a screenshot of it before it disappears. That is your evidence. If you find nothing, say so plainly and move on to the provenance check below. A negative result is a result, not a verdict.

Worked example: the Peter Obi photo that fooled a timeline

The test photo for this guide is a real, documented face swap. In late January 2023, a post on X claimed to show Nigerian presidential candidate Peter Obi sitting at a table with two bottles of Hero beer, a brand associated with the Biafran separatist movement. The post was shared by a rival party supporter, and Africa Check puts the tweet’s views at more than 737,000. The photo was fake. Both AFP Fact Check and Africa Check debunked it within days, and their findings give us a perfect test case: a swap that fooled hundreds of thousands of people, with the original photo still online to compare it against.

Here is what the fact-checkers found. The original photo, posted on 23 January 2023 by a user called JayNaija, shows a man eating amala, a yam flour meal, on a date. To make the fake, the man’s head was replaced with Obi’s, his orange T-shirt was recoloured yellow, and the two energy drink bottles on the table were replaced with Hero beer bottles. AFP confirmed the match with four details that survive in both versions: the angle of the right arm is the same, the sauce bottles sit where the beer bottles now sit, the food bowls are identical, and the same man and water bottle appear in the top-left corner. The swap itself shows in the details AFP flagged: a halo edge, a band of lighter pixels, around the beer bottles and around Obi’s head and neck, and a head that is out of proportion to the rest of the torso.

I ran the same checks on the photo myself on 15 November 2026, using a 245 by 321 crop of the altered photo taken from AFP’s own comparison graphic. The file carries no EXIF data, because it is a screenshot of a graphic, not a camera original. Here is what each tool returned.

CheckWhat I didWhat it returned
TinEyeUploaded the cropped altered photo1 match across 86.7 billion images, earliest indexed 26 April 2023, an Africa Check page hosting the file ObiBeer_Fake.png with a FAKE stamp on it
Google LensUploaded the same cropNo numeric total shown. 13 exact matches listed. Top result: the AFP Fact Check page of 8 February 2023, followed by Facebook and Nairaland reposts
Hive demoUploaded the same crop, no sign-up0% likely AI-generated image, 0.5% likely deepfake. Verdict: not likely to contain AI-generated or deepfake content. No Content Credentials found

Read those three rows together, because that is the lesson of this article. The reverse searches worked. TinEye led straight to the fact-check image, and Lens put the AFP debunk at the top of its exact matches. The photo’s history, not its pixels, gave it away. The detector, meanwhile, missed it completely. Hive scored a real, documented face swap at 0% likely AI-generated and 0.5% likely deepfake, and called it not likely to contain AI-generated or deepfake content. The swap is a small region inside a real photograph, so a whole-image detector has almost nothing to detect. If you had trusted that score alone, you would have called a proven fake genuine.

Two cautions belong with these numbers. First, the TinEye match is the fact-checkers’ own copy of the image, not the original JayNaija photo. Reverse search finds where an image has been published, and here it published the debunk first. That is still a win: the top result tells you the photo is fake. Second, the Lens results include noise. Several of the 13 exact matches are reposts on Facebook and Nairaland that have nothing to do with the original photo. Count the results, then read them. A match list is only useful if you open the pages on it.

Why did an AI detector score a real swap at 0%?

Because it was asked the wrong question. A whole-image detector answers “was this entire image generated by AI?” and a face swap is not an entire generated image. The room, the table, the bowls, the bottles and the body are all a real photograph. Only the face was replaced. The detector looks for generation signals spread across the frame, and in a swap those signals cover a small patch of the picture, if they survive at all.

Compression finishes the job. The photo in our worked example had already been screenshotted, cropped and re-uploaded several times before I tested it. Each save strips a little forensic detail. Bellingcat’s 2023 test of a leading detector showed the same effect: after social-media-level compression, the detector missed 7 of 10 AI images. A swap that starts out subtle becomes invisible to a detector after a few shares.

That does not make detectors useless. It makes them one layer, not the verdict. Use them the way our guide to spotting AI-generated images explains: as a signal to weigh with the visual checks, the reverse search and the provenance check. A high score is a warning. A low score is not clearance. Any tool that gives you a single percentage and stops there is selling you certainty it cannot deliver.

What do Content Credentials prove, and what do they not prove?

Content Credentials, the C2PA standard, are a signed record attached to a file that says where it came from and what was done to it. When a photo carries valid credentials, you can read its history: which camera or app made it, whether AI tools edited it, and whether the file has changed since. Our guide to checking a photo’s Content Credentials walks through the read step by step, and the official verify tool at verify.contentauthenticity.org reads them for free.

The trap is reading too much into an empty result. Most photos have no credentials at all. Social platforms strip them on upload, screenshots never carry them, and a swap made in an app that does not sign its output leaves no trace. In our worked example, the Hive demo reported no Content Credentials found for the swapped photo, and that tells you nothing either way. It is the normal state of a photo that has been around the internet for three years.

So use the rule this way. Credentials present and valid are useful evidence, because a signed history is hard to fake. Credentials absent prove nothing. A photo with no credentials can be genuine, and a photo with no credentials can be a swap. If the file you hold does carry credentials, read them before you trust the pixels. If it does not, fall back on the reverse search. That is the check that does the work when provenance is gone.

What should I do if a swapped photo shows me?

Do not share it on, even to warn people, because every share spreads the face further. Save the evidence first: the photo itself, the page it appeared on, the account that posted it, and the date you found it. Then report it where it is posted. Every major platform has a report route for manipulated media of a real person, and our guide to reporting an AI-generated image of you walks through the platform steps.

The law has caught up faster than most people realise, but only for intimate images. In the United States, the TAKE IT DOWN Act, signed on 19 May 2025, makes it a crime to knowingly publish someone’s intimate images without consent, including AI-made ones, and since 19 May 2026 covered platforms must remove a reported image within 48 hours of a valid request. The Federal Trade Commission enforces that duty, and complaints go through TakeItDown.ftc.gov. In England and Wales, sharing a non-consensual intimate image, including a deepfake, is an offence under the Sexual Offences Act as amended by the Online Safety Act 2023, and creating one became an offence on 6 February 2026 under the Data (Use and Access) Act 2025. This is general information, not legal advice, and the rules differ by country and by what the image shows.

For a non-intimate swap, the route is the platform report plus the record you kept. A swapped photo used to damage your reputation is a defamation and impersonation problem, not a removal-form problem, and a solicitor can advise on that far better than a web guide. What you should never do is confront the poster or accuse the person whose face was used. In romance scams, the person in the photo is usually an innocent third party whose pictures were stolen. The Associated Press reported in November 2026 on Hunter Bigham, a Mississippi man whose photos were used in 177 fake Facebook profiles over eight years, with victims who contacted him reporting about 1.4 million dollars in losses. He was the bait, not the scammer. Aim your report at the account that posted the swap, not the face in it.

Video swaps follow the same idea but need different checks, because motion gives you extra tells and takes some away. If the swap you are worried about moves, use our guide to checking whether a viral video clip is AI-generated, which covers the frame-by-frame method. Live video calls are a separate problem again: a swapped face on a call is usually paired with an urgent payment request, and the defence is to verify on a second channel you started yourself, not to keep staring at the screen.

What mistakes make a face swap check fail?

The most common mistake is trusting one check. A clean TinEye result, a 0% detector score, or a photo that looks fine at a glance each feel like an answer, and none of them is one. Run the layers. If the visual checks say swap and the detector says clean, trust the visual checks and the reverse search. The detector is the weakest layer for this exact kind of fake.

The second mistake is checking the wrong copy. A screenshot of a photo is a new file with new compression, and the seam you need may already be smoothed away. Always work from the best copy you can get, and if the photo arrived inside a chat app, ask for the original file before you judge it.

The third mistake is accusing the wrong person. A swapped face belongs to someone who may have no idea their photo was used. Do not message them, do not name them publicly, and do not treat a match as proof that the person in the photo did anything. The match proves the photo is public. Nothing more.

The fourth mistake is the opposite one: deciding a photo is real because no tool flagged it. A swap that survives every check still exists, and the competitor guide that sets the bar for this topic says so plainly. When the stakes are high, a photo that passes every pixel check still has to pass the context check: who sent it, where it first appeared, and whether any reputable source carries the same moment from a second angle. If those answers are thin, treat the photo as unverified, however clean it looks.

FAQs

Can I spot a face swap just by looking at a photo?

Sometimes, but not reliably. A rough swap shows at the hairline, the jaw and the ears, and bad lighting gives it away fast. A good swap survives a casual look, especially on a phone screen. Zoom to full size, check the boundary and the light, then reverse search the photo. Your eyes are the first check, never the only one.

Will a reverse image search always find the original photo?

No. A reverse search finds the original only if that photo is published and indexed somewhere. A new photo, a heavy crop or a flipped copy may return nothing at all. A zero on TinEye or Google Lens never clears a photo. It only means the search found nothing this time, so run the other checks too.

Why did an AI detector score a real face swap at 0%?

Because a swap is a small edit inside a real photograph, and whole-image detectors look for generation signals across the whole frame. Most of the photo is genuine, so there is little for the detector to find. Compression from re-uploads strips even more signal away. A low score means the detector found nothing, not that the photo is real.

Does a photo with no Content Credentials mean it is fake?

No. Most photos online carry no Content Credentials at all, because platforms strip them on upload and screenshots never have them. A photo with no credentials can be genuine, and a swapped photo can also carry none. Valid credentials are useful evidence when they exist, but their absence proves nothing either way.

Can I use this check on a dating profile photo?

Yes, and dating profiles are one of the best places to use it. Reverse search the profile photo, then check the ears, the hairline and the light at full size. A swapped or stolen photo often appears on other profiles or stock sites. If the photo fails any check, do not send money or personal details, and report the profile.

What should I do first if someone swapped my face into a photo?

Save the evidence before anything else: the photo, the page it appeared on, the account that posted it, and the date. Do not share it on, even to warn people. Then report it on the platform where it appeared. For intimate images, use the platform report and the removal routes in our reporting guide, and keep the original files safe.

Ali Butt is the lead writer and SEO specialist at Image Search Techniques, where he covers reverse image search, visual search, and online image verification. With over four years of experience in SEO and content, he writes practical, hands-on guides that help readers trace a photo to its source, identify people, products, plants and brands from a picture, protect their own images online, and spot catfish, fakes and AI-generated images across Google, Yandex, Bing, TinEye and more.

Leave A Reply