How to Spot Deepfakes Online? | 5 Surefire Strategies

Viruses, chat bots, and unsolicited emails seem to be on the bench as we are facing a new kind of digital threat — deepfakes. These already sophisticated pieces of synthetic media are becoming harder to detect by the day, deceiving us more effectively and contributing to a sense of virtual paranoia we haven’t experienced since the dawn of broader artificial intelligence (AI). 

Yet, underneath all the surface refinement made possible by advances in deep learning technology, there still lurks the ugly face of artifice that a well-trained eye can’t but notice. In this guide, we’re looking at five telltale signs of a deepfake and giving you the tools to help you spot them. 

Deepfakes: understanding digital deception

The dangers of deepfakes are plenty. In fact, ExpressVPN’s study on deepfakes shows that not even our memories are safe from this type of digital manipulation, as they can influence what we remember and our perception of reality. Considering the broad application of this new technology, it is becoming difficult to ignore its potential risks, both to individuals navigating the digital landscape and society.

As mentioned, the deepfake technology has gone far. Just in the last few years, we’ve gone from producing obvious, almost purely humorous pieces of digital media to dishing out not only photorealistic but hyperrealistic images that often seem more real than genuine audio-visual content. 

How do we protect ourselves, then, and how do we prevent the misuse of deepfake technology? The most obvious mindset to adopt is that of skepticism. By being wary of potentially deceptive content, we can contribute to an online atmosphere intolerant of fraudulent and malicious information.

How to spot a deepfake?

Detecting deepfakes requires a detailed approach that takes into account a number of factors. In other words, it is wise to rely on more than one strategy and use all tools at your disposal if you wish to be sure you are dealing with a deepfake rather than a legitimate image or recording. For starters, you can:

1. Look at the face and the eyes

Unnatural eye and facial movements are still the most obvious warning sign of a deepfake. Despite the steady advancements in deep learning technology, it is still impossible to replicate some things faithfully, for example, blinking. When we speak to or address someone, we tend to maintain eye contact or follow them as they move, which is still tough to mimic with the current deepfake models.

If you become suspicious, look for glitches, distortions, and choppy shifts in eye movement. Usually, if we are indeed dealing with a deepfake video, something will be out of sync with the rest of the footage. 

The same goes for facial expressions and movements. Namely, the feeling of the “uncanny valley” is often inevitable with deepfakes, as replicating natural emotional reactions that follow free-flowing natural speech is still in the realm of imagination. 

That is especially true when individuals in the footage move rapidly or switch positions. When that happens, it is usually likely that their nose, eyes, or other facial features will either be distorted or fail to move in tandem with each other.

2. Analyze the lighting

Focusing too much on minutia, most people engaged in deepfake forensics used to neglect lighting as a potentially sure indicator of artificiality. 

As with facial features and expressions, we need to look for inconsistencies. As deepfakes are usually the result of an image superimposed onto an already existing one, it is likely that the lighting will not be the same in both layers. Usually, the mismatch is most obvious when it comes to shadows or unnatural highlighting around the face that often does not match that of the environment surrounding it. 

3. Check out the clothing, hair, and accessories

The quickest way to create a deepfake is to replace an individual’s face and leave the rest as it is. While sloppy, this method is sufficient to trick a number of unwary or inexperienced internet users. However, once you start paying attention, you’ll notice all the drawbacks of this crude deepfake design.

For one, clothing and hair may appear unnatural or out of sync with the rest of the scene. There might be wrinkles in the fabric or wrinkles that don’t really match the person’s movement, and their hair might be oddly styled or, for example, disheveled by the wind in a scene where there is no wind.

In some instances, clothes, hair, and accessories such as jewelry can behave really strangely. For example, they might float, jut out at odd angles, or exhibit other forms of inconsistencies that make them stick out from the rest of the scene.

4. Listen to the voice

Relying on voice alone is probably the hardest deepfake detection strategy, as the technology has advanced the most in this regard. Nonetheless, there are still some indicators that will help you distinguish between fake and authentic voices.

First, deepfakes are more likely to feature unnatural, robotic speech patterns. Listen for flaws or oddities in cadence, rhythm, pitch, and voice modulation. Whenever you can, note any inconsistencies in them and then compare them with recordings you know are genuine.

Next, try pinpointing down any uncharacteristic speech traits. Each person has a unique voice, accent, and vocal mannerisms. Those who are used to hearing someone can often intuitively recognize synthetic voices if some of their qualities don’t match those of the person whose speech they’re supposed to replicate. 

Finally, it also helps to pay attention to background noises. For example, there should be no wind noises in recordings meant to be taken indoors. Likewise, if there are traces of other (poorly edited) voices following the main voice line supposed to be the only one in the recording, the chances are that the recording is fake.

5. Seek expert opinion

When unsure, it helps to look for expert opinions. There are people specializing in detecting deepfakes and developing software that facilitates such endeavors. With analytics tools at their disposal, they can analyze all audio-visual artifacts and detect anomalies not visible to the naked eye.

Context is also important when detecting deepfakes, and it is often the deciding factor when determining what is fake and what is not. Those well-versed in AI technology will know where to look for sources, how to figure out the motivation for the deepfake, and what kind of social context it originated in, all of which can provide them with the additional insight necessary to uncover the truth.

How can deepfakes pose a threat to us?

While a lot of people agree deepfakes are a threat, their exact reasoning is not always the same, and the question remains of whether deepfakes are deceptive in essence and therefore unacceptable, or are they harmless, with the real danger lying in the way we use them?

Whatever the case may be, the reality is that it is very likely that we’re going to reach a stage where about 90% of online content is AI-generated. If we are seeing such a surge in fraudulent and malicious behavior now, one is excused for expecting further violations of personal integrity, identity theft, and digital manipulation down the line. 

Only time will show whether we’re dealing with a slippery slope fallacy or whether our worries are justified. Still, it is better to stay prepared. After all, combating potential threats posed by deepfake technology starts with the right mindset, which, combined with additional technological safeguards and increased digital literacy, helps foster a more secure virtual environment.

Disclaimer: The content on this site should not be considered investment advice. Investing is speculative. When investing, your capital is at risk.’

FAQs on how to spot a deepfake