🪞 Online reputation & deepfakes

AI-generated or altered images: how to verify them

Retouched photos, swapped faces, invented scenes: a practical method to verify a suspicious image through visual details, reverse image search and context.

Photo manipulation once required skill and hours of work. Today a realistic image of a person in a place they have never been, or with one face swapped for another, can be produced with a few text prompts. For anyone's reputation this changes a lot: a fake photo can circulate for days before someone asks whether it is genuine. Learning to verify it is an increasingly useful skill, both when the image concerns you and when you are about to share it.

Three types of image to tell apart

Not all fake images are alike. It helps to distinguish:

Type Example What to look for
Fully generated A scene that never existed Inconsistent details, no source
Partly altered A face inserted into a real photo Edges, lighting, proportions
Genuine but out of context A real photo with a false caption Original date and place

The third case is perhaps the most common and requires no artificial intelligence at all: you just take a real photo and give it a different meaning.

Look at the details

Generated images have made enormous progress, but zooming in calmly on certain spots often reveals errors:

  • hands and fingers: the wrong number, fused fingers or impossible joints;
  • text and lettering: signs, labels or number plates with distorted letters or words that do not exist;
  • accessories: mismatched earrings, glasses blending into the skin, asymmetrical jewellery;
  • background: people in the background with undefined faces, buildings that could not stand, lines bending for no reason;
  • light and shadow: shadows pointing in different directions, reflections in the eyes that do not match the scene;
  • skin and hair: an overly smooth look, strands dissolving into the background.

In partly altered images, focus on the borders of the edited area: an inserted face often has lighting, sharpness or grain that differs from the rest of the body.

Reverse image search is the most powerful tool within everyone's reach. You upload the photo, or paste its address, into a search engine that supports this feature and get the pages where the same or similar images appear.

This check can reveal:

  • the original photo, perhaps with another person or in another context;
  • the earliest date the image appeared online, useful for exposing old photos passed off as recent;
  • different versions, showing where it was edited.

Try several search engines: results vary a lot between them. If the image has been cropped or mirrored, also try searching a cropped detail, such as the background or an object.

Question the context

As with videos, context is often decisive:

  1. Who published the image first? A recently created anonymous account deserves more doubt than a news outlet or an official profile.
  2. What is it supposed to prove? The more "perfect" the image is for backing an accusation, the more it needs checking.
  3. Are there other photos of the same moment? A real event with several people present usually produces more images from different angles.
  4. Are the details consistent? Season, clothing and recognisable places must match what is claimed.

The data hidden in the file

Photos taken with a smartphone often contain extra information, such as the date taken and the device model. It can be useful, but with an important caveat: social networks usually strip it on upload and it can also be edited. Its presence or absence, on its own, proves nothing.

Some generation tools and some platforms are starting to add labels or information about the origin of content. If you find them, they are a useful clue; if they are missing, do not conclude that the image is authentic.

If the fake image concerns you

Save the image and take a screenshot of the page with the address visible. If you find the original photo through reverse search, save it together with the address where it appears: it is the most convincing way to show the manipulation. Then flag the content to the platform, explaining in one line that the image is altered and pointing to where the original can be found.

In short

  • Distinguish between generated images, altered ones and genuine ones taken out of context.
  • Zoom in on hands, lettering, accessories, background and the borders of suspicious areas.
  • Reverse image search can reveal the original photo and its earliest date.
  • Ask who published it first and what it is meant to prove.
  • If it concerns you, save the original you find: it is the clearest proof of manipulation.
MM
Marco Marzaduri

Marco Marzaduri works in cybersecurity, digital forensics and web development. With degrees in computer science and psychology, he combines technical analysis with an understanding of the human factor.

Read also