close
close

How to identify AI-generated videos

Sorry to disappoint, but if you're looking for a short list of easily identifiable and foolproof methods to detect AI-generated videos, you won't find it here. Gone are the days when AI Will Smith grotesquely ate spaghetti. Now we have tools to help you create compelling, photorealistic videos in just a few clicks.

Currently, AI-generated videos are still a relatively young modality compared to AI-generated text, images, and audio, as getting all the details right is a challenge that requires a lot of high-quality data. “But there is no fundamental barrier to obtaining higher quality data,” just labor-intensive work, said Siwei Lyu, a professor of computer science and engineering at the University at Buffalo SUNY.

This means you can expect AI-generated videos to get much better very soon, eliminating the telltale artifacts – errors or inaccuracies – like morphing faces and shape-shifting objects that characterize current AI creations. So the key to identifying AI-generated videos (or any AI modality) lies in AI competency. “I understand that [AI technologies] are growing and the core idea of ​​“Something I see could be generated by AI” is more important than, say, individual clues,” said Lyu, the director of UB’s Media Forensic Lab.

Navigating the AI-infested web requires online savvy and good judgment to recognize when something is wrong. It's your best protection against being fooled by AI deepfakes, disinformation, or just plain low-quality garbage. It's hard to develop this skill because every aspect of the online world fights it to get your attention. But The good news is that it is possible to optimize your AI detection instincts.

“Through learning [AI-generated images]“We believe people can improve their AI skills,” said Negar Kamali, an AI researcher at Northwestern University’s Kellogg School of Management who co-authored a guide to identifying AI-generated images. “Even if I don’t see any artifacts.” [indicating AI-generation]my brain immediately thinks, 'Oh, something's wrong,'” added Kamali, who has studied thousands of AI-generated images. “Even if I don’t find the artifact, I can’t say for sure that it’s real, and that’s what we want.”

What to Watch for: Imposter Videos vs. Text-to-Image Videos

Before we can dive into identifying AI-generated videos, we need to distinguish the different types. AI-generated videos are generally divided into two different categories: impostor videos and videos generated by a text-to-image diffusion model.

Scam videos

These are AI-edited videos that consist of face swapping – where one person's entire face is swapped for that of another person (usually a celebrity or politician) and made to say something wrong – and from a lip sync – where a person's mouth is subtly manipulated and replaced with different sound.

Imposter videos are generally pretty compelling because the technology has been around for a while and they build on existing footage rather than creating something from scratch. Remember the deepfake videos of Tom Cruise a few years ago that went viral for their persuasive power? They worked because the creator, Chris Ume, looked a lot like Tom Cruise, worked with a professional Tom Cruise impersonator, and did a lot of minute work, according to one interview with Ume from The Verge. These days there are a plethora of apps that can accomplish the same thing and even, shockingly, incorporate audio from a short sound snippet the creator finds online.

However, there are a few things you should pay attention to if you suspect an AI video deepfake. First, look at the format of the video. AI video deepfakes are typically “shot” in a talking head format, where you can only see the speaker’s heads and shoulders, with their arms not visible (more on that in a moment).

To detect face swaps, look for flaws or artifacts on the edges of the face. “Usually you see artifacts when your head moves at an angle to the camera,” said Hany Farid, a digital forensics expert and professor of computer science at UC Berkeley. As for the arms and hands: “If the hand moves or something obscures the face, [the image] “It will be a small mistake,” Farid continued. And pay attention to natural movements of the arms and body. “If you just see that” — on our Zoom call, Farid keeps his arms stiff and at his sides — “and that person isn't moving at all, that's a fake.

Destructible speed of light

If you suspect a lip sync, focus your attention on the person's mouth – especially their teeth. In fakes, “we saw people who had irregularly shaped teeth,” or the number of teeth changed over the course of the video, Lyu said. Another strange sign to look out for is “bottom face wobbling,” Lyu said. “There's a technical process where you have to precisely match that person's face,” he said. “As I speak, I move my face a lot and the human eye can detect this orientation if you are just a little imprecise.” This makes the lower half of the face appear more fluid and rubbery.

When it comes to AI deepfakes, Aruna Sankaranarayanan, a research associate at the MIT Computer Science and Artificial Intelligence Laboratory, says her biggest concern is not deepfakes of the world's most famous politicians like Donald Trump or Joe Biden, but important figures who might not so well known. “Inventions that they come up with that distort certain facts, when most of the time you don't know what they look like or what they sound like, that's really hard to refute,” said Sankaranarayanan, whose work focuses on political deepfakes. This is also where AI expertise comes into play; Videos like these require some research to verify or debunk.

Text-to-image videos

Then there are the sexy newcomers: the text-to-image diffusion models, which generate videos from text or image prompts. OpenAI made a big splash with the announcement of Sora, its AI video generator. Although it's not available yet, the demo videos were enough to amaze people with their level of detail, vivid photorealism, and smooth tracking, all supposedly through simple text prompts.

Since then, a number of other apps have popped up that let you turn your favorite memes into GIFs and imaginative scenes that look like they were made by an entire CGI team with a Disney budget. Hollywood creatives are rightly outraged by the emergence of text-to-image models that probably trained in their work and are now threatening to replace it.

But the technology isn't quite there yet, as even those Sora videos likely required sophisticated and time-consuming editing. Sora's demo videos consist of a series of quick edits because the technology isn't yet good enough to create longer, error-free videos. Here’s how to be especially wary of short clips: “If the video is 10 seconds long, be suspicious. There’s a reason why it’s short,” says Farid. “Basically, with text-to-video, you can't have a single cut that's a minute long,” he continued, adding that this is likely to improve over the next six months.

Farid also said to pay attention to “temporal inconsistencies,” such as “the building added a floor or the car changed color, things that are not physically possible,” he said. “And often this happens away from the center of attention.” So focus on the background details. You might see unnaturally smooth or warped objects or change a person's size as they walk around a building, Lyu said.

Kamali recommends looking for “sociocultural implausibilities,” or contextual clues where the reality of the situation seems implausible. “You don’t see the signs right away, but you get the feeling that something is wrong – like a picture of Biden and Obama in pink suits,” or that Pope in a Balenciaga puffer jacket.

Context clues aside, the existence of artifacts will likely decline very soon. And Wall Street is ready to bet billion dollars on it. (That is, venture capitalism is not really known for cost-effective valuations of technology startups based on solid evidence of profitability.)

The artifacts may change, but good judgment remains.

As Farid told Mashable, “Come and talk to me in six months and the story will have changed.” So if you rely on certain clues to check whether a video is AI-generated, you might get into trouble.

Lyu's 2018 article on detecting AI-generated videos because subjects weren't blinking properly received widespread attention in the AI ​​community. As a result, people began to look for errors in eye blinking, but as technology advanced, so did more natural blinking. “People started thinking, if there is a good eye blinking, it must not be a deepfake, and that is the danger,” Lyu said. “We actually want to raise awareness, but we don’t want to commit to specific artifacts because the artifacts will change.”

Building awareness that something exists could AI-generated will “trigger a whole sequence of actions,” Lyu said. “Check who is sharing this? Is this person reliable? Are there other sources related to the same story and has this been confirmed in any other way? I think these are the most effective countermeasures against deepfakes.”

For Farid, identifying AI-generated videos and misleading deepfakes starts with where you get your information from. Take this AI generated images that was shared on social media following Hurricanes Helene and Milton. Most of them were obviously fake, but still had an emotional impact on people. “Even if these things aren't very good, that doesn't mean they don't penetrate, it doesn't mean they don't have an impact on the way people absorb information,” he said.

Be careful when getting your news from social media. “If the image feels like clickbait, it is clickbait,” Farid said, adding that it all comes down to media literacy. Consider who posted the video and why it was created. “You can't just look at something on Twitter and say, 'Oh, that must be true, let me share it.'”

If you're suspicious of AI-generated content, check other sources to see if they're also sharing it and if everything looks the same. As Lyu says, “A deepfake only looks real from one angle.” Look for other angles of the instance in question. Farid recommends websites like Snopes and Politifact that debunk misinformation and disinformation. As we all continue to navigate the rapidly changing AI landscape, getting the work done – and trusting your gut – will be crucial.

Topics
Artificial intelligence