How-To News

How to Spot Deepfakes

By Samuel Cardwell / Dec 9, 2023

There are a variety of techniques that can be used to help you spot deepfakes, from quick tips to help you examine it yourself to software and extensions which can do it for you. Starting with some common giveaways of a deepfake, this article will act as a guide through three different techniques for spotting fabricated imagery from real content.

Spotting Deepfakes Yourself:

Visual deepfakes are frequently inconsistent and imperfect. As a result, a quick analysis of the imagery can highlight mistakes and inconsistencies to suggest whether the content is real or fabricated. When trying to spot deepfakes, some common things to keep an eye out for include:

  • Inconsistency in lighting or reflections e.g reflected lights in the eyes, windows or shadows
  • A strange background which doesn’t make sense Often AI generated images leaveweird shapes or colours in the surroundings.
  • Content quality that does not quite make sense. Mistakes made when fabricating content can be more easily hidden in lower resolution imagery or audio with higher levels of background noise, meaning that deepfakes can be made more convincing by presenting them in a lower quality.
(image of a deepfake where the background contains a weird attempt at another person)
If you focus on the face, it can be easy to ignore the weirdness of the background
(image of a face where the background is a weird pattern, and on either side of the subject there are strange random features)
Once again, the features of the people in the background are completely wrong
(image of a deepfake where everything looks real, but the background is a weird red and white pattern which does not make much sense)
It is worth thinking whether the background makes sense for a normal image.

When the imagery contains people, it is important to focus on the face as this is the most likely place to be doctored by deepfake technology. When examining the face, pay attention to:

  • Features that are positioned incorrectly, or appear to move strangely
  • A blurry face with skin that looks too smooth or wrinkly, or perhaps the wrong age compared to other features
  • Pay close attention to the mouth as often audio will not quite match its movements
  • The eyes often look strange or unnatural, potentially blinking too little or too much
  • Often, the eyes can move around strangely and not look around in co-ordination with each other
  • The edges of hair looking strange or blurred
  • Teeth that don’t make sense
an image of a deepfake where one eye looks fabricated and abnormal
The eyes in this deepfake don’t look quite right
image of a deepfake where the subjects teeth are erratic and inconsistent
The teeth in this deepfake do not make sense
image of a deepfake where the skin looks oddly wrinkled and blurry
The skin looks slightly odd or out of place in this deepfake

Some other common indicators of a deepfake are:

  • Inconsistency in clothing or accessories
  • Strange glasses or other glass objects
  • Hands with the incorrect number of, or just weirdly shaped, fingers
image of a deepfake where the glasses have made the eyes appear swollen and unnaturally misshapen
Here, the deepfake has struggled to produce convincing glasses, messing up the eyes.
image of a deepfake where the subject only has an earring on one ear
In this deepfake, the subject only has one earring.

Using Verification Principles to Identify Deepfakes:

There are a variety of advanced techniques that experts use to verify digital content, but instead of attempting to cover all of these in one article it is worth discussing the basics to help you begin to spot deepfakes. It should be noted that these tips are not infallible rules which can provide certainty over whether a piece of content is fabricated or not. Instead, they can point you in the right direction and give a general sense of credibility and the likelihood of authenticity.

  • Does the content make sense contextually? E.g people saying things out of character, crazy sounding headlines or the marketing of bitcoins and tech products.
  • Has anyone else verified the content? Check the comments or reposts of something on social media, or do a quick google search to see if the content seems to be real
  • Where is the content coming from? Is it a reliable news source, celebrity figure or channel that is posting the content. Does the account have a history of disinformation?

A surprising amount of deepfakes can be identified by simply applying a certain level of scrutiny and common sense.

Software for Spotting Deepfakes:

A lot of the more powerful deepfake detection software is only available for commercial use, leaving just a few solid options for individuals to use to spot deepfakes

DeepFakeProof is a browser extension that will automatically analyse images that are loaded on sites that you visit and try to determine whether they are deepfakes or not. In our tests so far, it has been quite accurate but is limited in that it cannot analyse video content.

InVID is a browser extension that is pitched as a toolkit for general verification. Incidentally, this means that it also has some useful tools for spotting deepfakes. For example, you can use the magnifying tool to zoom into parts of an image to look closer for any of the inconsistencies mentioned above. You can also do a forensic check which will spot artefacts in an image which might suggest it has been doctored. However, this cannot tell you exactly what changes have been made.

If all else fails, a quick reverse image search to see where an image has come from can help indicate its authenticity.

Practicing Spotting Deepfakes:

To put what you have learned to the test, you can practice spotting deepfakes at some of the sites below:

Which face is real? – Great for quickly practicing identifying deepfake photos yourself (also the source of most of the deepfakes in this article)

DetectFakes – Good for practicing identifying deepfake videos yourself

Spot the deepfake – Includes a short introductory guide to help learn more about how to spot deepfakes

Tags: , , , ,