Sorry for the disappointment, but if you're looking for a quick list of innocent ways to detect video generated by AI, you won't be able to find it here. There's no longer a time when I eat spaghetti with Ai Will Smith Grotesquely. Yes, there are a few tels, but AI video makers are always getting better, and the latest tools can create compelling light field registration videos with just a few clicks.
Currently, AI-generated videos are relatively early modalities compared to AI-generated text, images and audio, as getting all the details correctly is a challenge that requires a lot of high quality data. “But there is no fundamental obstacle to getting high quality data,” said Siwei Lyu, professor of computer science and engineering at the University of Buffalo Sunny, “but there is no fundamental obstacle to getting high quality data,” only labor-intensive work said.
Over the past six months, AI video generators have been so good at creating realistic videos that they often take away casual scrollers. Telltale artifacts used to keep the game away, such as morphing faces and shape-shifting objects, are much more common. There is little evidence in the viral AI-generating videos of street interviews conducted on emotional support kangaroos, trampoline bunnies, or Google's VEO 3 models (the video can generate sound).
Like AI modalities, the key to identifying video generated by AI lies in AI literacy. “Understand that [AI technologies] Lyu, director of UB's Media Forensic Lab, said:
This tweet is currently unavailable. It may have been loaded or deleted.
To navigate the AI Slop Infested Web, you need to use online savvy, good judgment to realize that something can turn off. The best defense against being overdone by AI deepfake, disformation, or just low quality junk. It is a difficult skill to develop as every aspect of the online world fights it for your attention. but The good news is that you can fine-tune your AI detection instinct.
“By studying [AI-generated images]Negarkamari, an AI research scientist at the Kellogg School of Management at Northwestern University, co-authored a guide to identifying AI-generated images, said: [indicating AI-generation]my brain immediately thinks, “Oh, something's off,” Kamari added.
Notable: fraudulent videos and text-to-image videos
Before identifying the video generated by AI, let's distinguish between the different types. Videos generated in AI are generally split into two different categories: fraud videos and videos, which are generated by intertext spreading models.
Imposter Videos is an Ai edited video consisting of face swapping, letting the whole face is exchanged for another person (usually a celebrity or politician), and the person's mouth is subtly manipulated and replaced by different audio.
Scammer Video: Why regulators are cracking down
Scammer videos are generally quite convincing. This technology has been around for a long time, and is built from existing footage rather than generating something from scratch. I remember the Tom Cruise Deepfake video from a few years ago. Creator Chris Um looked a lot like Tom Cruise and worked with professional Tom Cruise impersonators, so they worked. Interview ume, via Virge.
These days there are plenty of apps that can achieve the same thing, and even horrifyingly, include audio from short soundbites the creators find online.
That said, if you suspect deepfark in AI video, there are a few things to look for. First, look at the video format. Deepfakes in AI videos are usually “shooted” in a talking head format. Here you can see the speaker's head and shoulders, and your arms will be invisible (more details are provided).
To identify face swaps, look for defects and artifacts around the face boundary. “When the head moves diagonally to the camera, the artifacts are usually displayed,” said Hany Farid, a digital forensics expert and a professor of computer science in Berkeley, California. Regarding arms and hands, “If your hands are moving, or something blocks your face, [the image] “Farid continued. And look at his arms and body for natural movement. “If this is all you're looking at” – our zoom call, Farid keeps his arms stiff and on his side, “And it's not moving at all, it's fake.”
If lip synchronization is suspected, pay attention to the subject's mouth, especially the teeth. With fake, “We've seen people with irregular teeth shapes,” or the number of teeth changes throughout the video, Lyu said. Another strange sign to look at is the “wobble of the lower half” of the face, Ryu said. “There are technical procedures that require that the person's face be accurately matched,” he said. “When I'm talking, I'm moving my face a lot, and if you're a little inaccurate there, the human eye can convey.” This gives the lower half of the face a more liquid rubbery effect.
Masculine light speed
This tweet is currently unavailable. It may have been loaded or deleted.
When it comes to AI Deepfakes, Aruna Sankaranarayanan, research assistant at the MIT Computer Science and Artificial Intelligence Institute, says her biggest concern is not the deepfakes of the world's most well-known politicians like Donald Trump and Joe Biden, but not the lesser-known and important figures. “It comes out of them and distorts certain facts. When you don't know what they look like or what they sound like most of the time, it's really hard to disprove,” said Sankaranarayanan, whose work focuses on the deep political things. Again, this is when AI literacy comes into play. Videos like this require some research to verify or expose.
In April 2025, Congress passed the Take it Down Act to share intimate images of federal crime or inauguration or non-consent. Another bill, called the No Fakes Act, passes through the Senate. It is intended to provide AI with legal protection against replicas generated.
How to find video from text to image
Text-to-image generators exploded in popularity while regulators crack down on fraudster videos. Now you can generate AI videos directly within ChatGpt and Google Gemini. Luma, Kling, and Freepik are just some of the other options of simple video generators that have grown online.
With short textual descriptions, you can generate any kind of video that dreams of your imagination. Most of the AI-generated videos shared online fall into the category “Hey, see what you can do with this cool new technology.” This ranges from absurdity like a cat jumping off an Olympic diving board to something really misleading, like a fake video of hurricane damage. But it all contributes to a confused, dystopian experience. There it becomes even more difficult to separate the fiction generated by AI from reality.
Furthermore, many accounts that distribute AI-generated videos benefit from ClickBait by intentionally deceiving users. With Tiktok, it is virtually impossible to know if AI-generated creators selling the latest skincare products. Videos generated to AI made with Tiktok's tools are automatically disclosed, but that does not prevent users from uploading videos generated or edited to AI made with tools outside the platform.
Experts say you can try and find clues to the context. Farid said he would focus on “temporal contradictions” in which buildings added stories or cars changed colours or physically impossible. “And often, where it's happening is far from the center of attention.” So hone the background details. Walking around unnaturally smooth or distorted objects, or buildings could change the size of a person, Ryu said.
This tweet is currently unavailable. It may have been loaded or deleted.
AI actors and deepfakes have not come to YouTube ads. They are already here.
Kamali says they look for clues of “sociocultural intolerance” or context in which the reality of the situation seems plausible. “You don't see Tertares right away, but you feel something's off – like the image of Biden and Obama wearing pink suits” or Balenciaga's Puff Jacket Pope.
The artifacts may change, but good judgment remains.
However, relying too much on certain cues to see if your video is generated by AI can lead to trouble.
Lyu's 2018 paper on the detection of video generated by AI in the AI community because subjects did not flash properly. As a result, people began to look for eye-opening flaws, but as technology progressed, they also had more natural blinking. “People have started to think that if they have good blinking eyes, it shouldn't be a deepfake, which is dangerous,” Ryu said. “We actually want to raise awareness, but we don't latch certain artifacts because they fix them.”
Recognize something Maybe When generated in ai, “the entire series of actions are triggered,” Lyu said. “Who is sharing this?
This tweet is currently unavailable. It may have been loaded or deleted.
In the case of Farid, identifying AI-generated videos and misleading deepfakes starts from where information is sourced. Take images generated by AI that have spread on social media in the aftermath of Hurricane Helen and Hurricane Milton. Most of them were obviously fake, but they still had emotional impact on people. “Even if these things aren't that good, it doesn't mean they won't penetrate, or it doesn't mean they won't affect the way people absorb information,” he said.
This tweet is currently unavailable. It may have been loaded or deleted.
Be aware that you get the news from social media. “If the image feels like ClickBait, it's ClickBait,” Farid said, all of it comes from media literacy. Think about who posted the video and why it was created. “You can't just look at something on Twitter and do something like, 'Oh, it has to be true, let me share it.' ”
If you're suspicious of AI-generated content, check out other sources to see if they're sharing them as well, and whether everything looks the same. As Lyu puts it, “Deepfakes look real only from one angle.” Search for other angles in the instance in question. Farid recommends sites such as Snop and Politifact. As we all continue to navigate the rapidly changing AI landscape, it will be important to do our job – and trust your gut.
How are AI companies labeled videos generated by AI?
Some AI companies, including Google and Openai, have a way to label AI-generated videos. With every video generated by VEO, Google incorporates an invisible watermark called SynthID. After the launch of VEO 3, after sparking a wave of concern, the company added a visible watermark to label it as AI generated.
Openai, Adobe, and other companies use technical standards developed by the nonprofit coalition for content origin and reliability (C2PA) to label AI-generated videos and images with invisible watermarks.
The visible watermark may seem like an obvious solution, but it can also be easily removed. And then there's the question of whether they are important. A study from Stanford University's Human-Centered AI (HAI) Institute recently discovered that visible labels that indicate AI-generated content “do not change its persuasiveness.” After all, we're used to all sorts of meaningless logos in viral videos. It's easy to tune them visually.
On the other hand, the invisible watermark is burned into the metadata. This makes it difficult to delete and makes tracking easier.
Standards like C2PA are a step in the right direction, but it is up to the company to voluntarily adhere to these standards now. Perhaps one day these standards will be enforced by regulators. In the meantime, our best bet is still healthy judgment and strong media literacy.
topic
artificial intelligence
