How Do You Know If A Video Is AI-Generated?
1st Mar 2024 Technology
3 min read
In the present world, statements like “Seeing is believing” or “I’ll believe it when I see it” are becoming less valid. And this is a result of the advent of deepfake videos generated by AI. With such videos, it is very easy to scam the recipients, unlike playing online blackjack in Canada.
At
first, algorithm-generated content was limited to pictures alone, but this has
now transcended into videos. These videos have rapidly developed over the years,
making it much more difficult to know when you can trust your eyes and when you
shouldn’t.
However,
regardless of how advanced artificial intelligence may be, there are still
certain tips that can point out the flaws of deep-fake videos. We shall take
this one after the other as we progress in this write-up. Dive in.
Watermarked Content
This
is the most basic means by which AI-generated videos give themselves away.
Usually, the algorithm watermarks such videos it
generates. For instance, looking at DALL-E2 videos, you are sure to find
watermarks in the form of five colored squares on every piece of content
downloaded from the website.
However,
this may not be valid in the long run because sometimes AI platforms allow
users to download the generated content without the watermark. Thus,
watermarking may not be a viable alternative to detecting AI videos.
Strange Shadows Or Light Flickers
As
a way of detecting videos created by machine algorithms, look out for strange
shadows, blurs, or even light flickers. This is because in such videos, light
or shadows may sometimes flicker on only the face of the person in view. And
this is very much true, especially when it’s a video gotten from low-quality
deep fakes.
Abnormal Body Language
Of
course, AI can never give off an illusion of fluency and seamlessness as the
actual human would. Thus, another way of detecting videos made from machine
learning is the body language. Algorithms usually find it difficult to
replicate humans, so if the body language is off, then it may be time to
question the credibility of the video.
An
example of such body language that may be off is the character not blinking at
all throughout the video.
Listen Closely
Audio
is another means by which AI-generated content can be detected. In the early
years of artificial intelligence, the sounds created were either disfigured,
distorted, or muffled, making it obvious that they weren’t natural. But in
recent times, that seems to have changed as synthetic sounds tend to come out
almost human-like.
However,
this doesn’t mean it is completely devoid of issues. One may notice choppy
sentences or even background sounds in some of those videos, thereby giving
them away.
Text on Clothing May Be A Giveaway
One
would expect that since it’s machine-generated, certain things, including text
on clothing, should be a given. But this is not so. Usually, with AI, the texts
on deep-fake videos are utter and indecipherable jargon. So, if you wish to
know if that image is fake, then zoom in and check it out.
Distorted Body Parts
This
one is common to both video and images created by algorithms. You can easily
notice deepfake videos through the unnatural or poorly drawn human features. This
is very common in the characters' hands, smiles, and eye gazes in such videos.
Sometimes,
you may notice people with six fingers or someone with two right arms and
hands. If so, then it’s a no-brainer that you’re dealing with a fake video.
Check The Source
Additionally,
in an attempt to determine how true a video is, it never hurts to trace the
source of the video. Although this may turn out to be difficult, especially if
it’s a viral video, it is essential to confirm if you’re looking at a fake.
Moreover,
if the video contains political or any other mainstream news, look into it and
see if other media outlets are reporting such news. If not, then it means the
video is questionable.
Use Detector Tools
In
this age and with the advent of AI-generated videos, there has been a launch of
tools to detect synthetic videos.
Such tools as Generative Adversarial Networks and GAN make use of algorithm
detectors to discover when generative content is present in videos. Although
the accuracy may not always be 100% in conjunction with all the other tips
listed above, you can do just fine with it.
The End
While
AI can bypass some of the above-listed giveaway features, it is impossible for
it to be 100% free of flaws. This is because, usually, at least one of the
visual details will have something lacking in authenticity. Therefore, in this
world of deep-fake videos, always feel free to question the credibility of
video content you see; good luck.