Deepfakes videos are AI-generated fake videos about people, places, or events that seem convincingly real but are in fact fake and machine-generated.
Deepfakes have made quite a stir on the internet. Some people are using these videos just for views and likes, while some are resorting to deception and political propaganda, stirring up chaos using fake videos.
Deepfakes have also been used for online scams. But they also make potent tools for blackmailing people.
Considering the fact that creating more and more convincing deepfakes is only becoming easier with the rise of AI tools, one of the useful skills you can learn right now is how to spot a deepfake video.
This simple guide explains how you can find and spot deepfake videos without scratching your head over what’s real and what’s not.
How to Spot a DeepFake Video: 10 Signs to Look for
Deepfake videos range from being easy-to-spot to hard-to-spot. There are many factors as to why that’s the case, and number one being that AI tools aren’t consistent at producing content.
Sometimes, with some specific prompts or instructions and/or engines, the tools generate more realistic deepfakes with less flaws than others.
This means that there will be videos that you can spot right away but some will take repeated, careful attempts to spot. But one thing is for sure, AI tools are largely imperfect at producing flawless videos as we speak.
While some people might go out of their way to edit the deepfake to cover the flaws and make it even harder to detect, we can assess telltale signs of deepfakes that can suggest whether the video is real or fake and draw a conclusion.
Following are 10 signs for spotting an AI clip. Let’s assign each sign 1 point that we count if it’s found in the video, with a maximum of 10 points.
The more points a video scores, the greater the chance its AI-generated.
Sign 1: Unnatural facial movements
One of the imperfections of AI tools is unnatural facial movements, like unnatural blinking of eyes, strange tensing of lips, or incorrect lipsing.
In less convincing AI videos, the facial movements are bluntly visible, which might just give its origin away and end your analysis of the video.
Look closely at how an individual’s face moves. Examine their facial features, including:
Eyes
Lips
Ears
Jaw
Eyebrows
Chin
Cheeks
For example, if the lips don’t match the words, eyes blink unusually, jaw seems disproportionate, or eyebrows change from thick to thin.
Sign 2. Strange skin texture
Most AI-generated images and videos have smooth-textured people in them. Their skin appears to be too smooth to be real, although it looks real if you don’t give it much thought.
Have you ever seen hyper-edited photographs of models or celebrities in which their skin appears near-flawless? That’s done through a method called retouching.
You can take that kind of smooth skin as a rough reference, though the skins appear much more smooth in the case of AI videos.
The overly smooth skins can also be described as somewhat blurry, without having normal skin features like pores and sometimes wrinkles.
Even if the faces have wrinkles on them, the wrinkles might also feel smooth. And if the faces have pores, they might look too even/smooth and unusually recognizable.
Sign 3. Lighting problems
Lighting is a complex phenomenon due to its nature in our physical world. If a video is real, it’s bound to have plausible lighting, both in the background and foreground.
However, luckily, for AI tools producing deepfakes, lighting is relatively easier to mess up, because it’s so intricate and nuanced, which can be a good sign of a deepfake.
For example, the light on a person’s face might not match the background or appear even on the rest of the body.
Sign 4. Odd eyes
One of the telling features whether a person is real or AI-generated is their eyes, because it is one of the particular human features these tools struggle to generate well and consistently.
In real videos, our eyes look and move naturally. The light reflections are natural and the eyes have no odd movements.
But in AI-videos, the eyes can appear strange. For example:
The pupils might feel like they’re changing size, or the white of the eyes might be getting mixed with the pupils and becoming black.
The person’s eye contact might as well be limited.
In some cases, people might not have dynamic and realistic pupil movements.
The eyes might also change form, becoming slightly smaller or larger.
The reflection of light in the eyes might not be accurate compared to the environment.
Sign 5. Audio mismatch
Another simple sign of AI videos is mismatching audios, compared to the people’s lipsing. The voices might not match the video’s settings, either.
For example, a deepfake restaurant, which should sound like a restaurant, might not do so satisfactorily; the audio might feel off and unnaturally more restaurant-type than it should normally be.
As for the speaker, their dialogue might sound robotic, both in terms of the material and the sound of their voice.
Their expressions and emotions might feel faked and stilted, and they might have unsynchronized lipsing.
Sign 6. Glitches, flickering, or blurs
Glitches and blurs are common in AI videos. However, good deepfakes might not have apparent glitches or blurs.
These might be negligible, but you might still be able to spot some if you pay attention.
For example, look for a strange flickering of pixels, blurry edges around the face, or parts of the image that look unreal.
Sign 7. Unnatural head movements
Head movements aren’t always flawless in deepfakes. A head moving stiffly or at an unnatural angle can also be a visual cue.
The difference might be too subtle and difficult to spot, but if it's one of the odds that seem to add up to the rest, that says something about the video.
Watch how the head moves compared to the rest of the person’s movement. Does it align as it should?
Sign 8. No or unreliable source
Not every clue has to be inside the video. A simple source-check might add to the likelihood and help decide if the video is a deepfake or not.
Trace the video back to its source. If the video’s source is unreliable, like a random social media account, chances are it's fake, even if it has a real, human-written caption.
The same goes if the video has no clear source of publication. It might be generated by an unknown person or group — all the more reasons to be careful and not trust the video.
Sign 9: High AI Detection Score
AI detection tools like Intel FakeCatcher and Norton Deepfake Protection (Norton Genie) can analyze suspicious videos and trace their AI patterns to check whether they’re real or not.
These tools can tell apart at least medium to intermediate-level AI videos that would require some human effort and attention to be spotted.
If a reliable tool reports high likelihood of AI being the video’s origin, it likely is, especially if the previous signs stack up.
However, a better way to use these tools is to use them as validators and not strictly as deciding factors, because these tools can be inaccurate sometimes.
Sign 10: Too many signs
As we said, designate each tell-tale sign a point and calculate how many points a suspected video scores.
If the video scores 8/10 points, it’s definitely a deepfake, in which case the points become 9/10, as this last point (“Too many clues") proves to be true.
But if a video scores, say, 1/10. There’s a small chance it might either be real or very well-edited (I mean extremely well-edited) deepfake, but the chances of the second possibility are super low.
In some cases, a single strong sign is enough to spot a deepfake. A strong sign is the one that’s obvious enough to distinguish fake from real and/or edited.
Tips for Spotting a Deepfake Video
In addition to the signs, here are some tips that will help you detect a deepfake video:
Check Comments: If you’re on a social media platform, checking comments can be helpful in spotting a deepfake video. Other people likely have already examined the video and figured out its origin. However, don’t blindly trust a few comments and make your own observations too, because, sometimes, people can judge peculiar yet real videos incorrectly. Also, check how many comments point out the video being AI. The more the unique comments, the more the likelihood of AI.
Watch Repeatedly: Watch the video more than once. It might take you more than several attempts to find flaws.
Slow the Video Down: If possible, slow the video’s speed to examine it more thoroughly. This will help you spot the visual transitions and possible glitches or abnormalities.
Watch Lots of AI Videos and Note the AI patterns: If you want to be able to better spot AI-generated videos, then try to get familiar with them. Watch lots of AI videos, uploaded on social media or web, and try to understand and get familiar with the peculiar AI patterns that feel common across the videos, such as unusual and smooth transitions and changing of limbs or body parts as well as all the signs listed above.
It’s better to doubt than trust
Remember, deepfakes are produced to deceive and make people believe the content. That’s why it's better to be careful and have doubts than to give in and trust.
So, if you come across a video that seems unusual but also realistic — almost flawless — scrolling past without forwarding is safer and better than liking and sharing, especially if the video is potentially controversial and involves a real, known person, like a celebrity, religious figure, leader, or a politician.
Conclusion
Deepfakes are fake yet convincingly real-looking AI-generated videos depicting people, places, or events.
You can spot a deepfake by examining it carefully and looking for the following 10 signs: Unnatural facial movements; strangely smooth-textured or blurry skin; unusual lighting: odd eyes; audio mismatching the video’s details; glitches, flickering, and blurs; unnatural head movements, the reliability of the video’s source, and the number of signs point to AI that stack up.
If you find a seemingly deepfake yet hard-to-detect video, it’s best to not trust it and scroll past just to be on the safer side.
Upgrade Now
