close
close

So you can see if a photo or a video is AI

I will always remember the time that my girlfriend, who was in her early 20, showed me a video that she believed that it trained Vladimir Putin in the gym.

Now most of us would not fall for a Putin training video. But since generative artificial intelligence (AI) becomes mainstream and advanced, it will be more difficult to say what AI is and what is not.

Recognizing is important. Especially in a day in which “we get so much of our information from social media, it is basically our main level to understand what happens around us or on the other side of the world,” said Siwei lyuDirector of the University of the Media Laboratory for Media Forensics in Buffalo.

“The world is so connected, but at any important event we cannot rely on audio-visual information from social media to tell us what happened.

AI-generated images and videos can be used to influence the results of political campaigns or our daily decision on buying products, said Lyu. Recently there were a number of ai-generated pictures online during the forest fires in Los Angeles, which caused incorrect ideas about what was really done on site.

What we see online can influence what we believe and how we formulate our opinions, Junfeng Yang, professor of computer science and co-director of the software system laboratory at Columbia University.

Lyu and Yang both agreed that we have to look at every photo or video online with the way of thinking that it could be generated with AI. This skepticism will give us a break before we go into the wrong conclusions.

How to recognize whether a photo or a video is generated with AI

One restriction: Generative AI develops and improves while we speak. Lyu and Yang warned that the treacherous signs they are now using to determine in the near future in the AI-generated content. The best way to recognize AI See and examine their source, both experts said.

Here you will find AI experts to determine whether a picture or video is initially generated by the AI. But according to Lyu, they don't rely too much on these signs. A lack of them does not mean that the picture or video is not AI.

Experts say there are some trademarks that prove that a picture or video is fake, but these elements could change if AI gets smarter.

Check the human hands.

According to Lyu and Yang, hands, arms and fingers don't look perfect.

“For example, if two people hug, their arms and hands can be in a very unusual configuration, they can turn together or their arms or fingers extremely long and so on,” said Lyu.

Make sure that the image respects real laws such as physics.

At first glance, an ai-generated picture captures the appearance and the feeling of a real world scene. However, the current generative AI models do not have the ability to replicate human anatomy, geometry and the laws of physics in their outputs, said Lyu. Current generative AI models still cannot really replicate how objects are connected to each other in the real world.

Watch the physical movements of people in the video, Yang advised. Of course, do you look how people would move in real life? Take a look at the lighting that falls on different objects in the picture or video. For example, if the light source seems to be from top left of the image, are the objects on this page correctly illuminated or are some illuminated from the right side?

While you scan after lighting lists, you can also scan of shade that are not logically sensible. If the light source appears from the picture from top left, are the shadows of the objects rightly pour on the opposite side (i.e. below right)?

Look into the eyes.

“With the human faces of AI-generated views, they look very realistic, but a way to say that they are generated with AI is actually looking into their eyes,” said Lyu. “The human pupil is a circular shape, almost like a perfect circle, but with A-generated faces have irregularly shaped pupils.”

Examine your mouth and teeth.

You have probably seen lip-synchronized AI-generated videos in which you match the mouth of a person to an audio clip you did not originally say. If you pause the video and slowly move the progress bar along the video time bar to inspect the video -ons frame at the same time, you may see that “inconsistencies in the number of teeth or the form of certain frontal teeth change from frame,” said Lyu.

According to Yang, the movement of the lips of the person can also appear unnatural.

“In the mouth of a real person you can also see the tongue and other structures, but for an ai-generated video, if the mouth is open, it is a dark, empty, free space,” noticed Lyu.

Listen to breath and natural breaks in the language.

“In the current voices with AI-generated voices, there is a lack of what we call paralingual properties,” said Lyu. “These are characteristics of human voices that have no meaning, such as breaks or breathing noises.”

As humans, we first have to take a breath in our lungs before we can lay out the air and move our vocal cords to create sound, Lyu explained. When we hear people speak, they can hear their breath and sometimes exhale. Currently in AI-generated voices “you do not have this breathing noise, so that you hear a very clean human voice without background noise,” added Lyu.

Be skeptical about things that look too perfect and are missing in detail.

“If you see a picture or video, that's too perfect or too surreal, this can be an indication of a generation of AI,” said Yang.

According to Yang, generative AI models are still challenges today to replicate all imperfect details and realistic texture of the objects and the background in the image or video.

For example, a human face or an animal fur may appear too smooth and almost with airbrush.

Scan yourself after repetition and duplication in the background.

“Sometimes in the video or in the picture you see the repetition of how this background person is correct, like 20 times and they all look similar,” said Yang. It can be a background person who duplicates several times to create a lot, or even a beach without a large texture variation between the sand.

Watch the flashing of the eyes when it comes to a video of a person.

Take a careful look at how a person blinks in the video, says Yang. A person in real life does not flash with a defined rate. When the flashing pattern is repeated, such as B. blinking every 10 seconds, it can be generated.

Check the source of the image or video.

If photos and videos with AI generated are becoming more and more realistic and the “quirks” mentioned above are remedied over time, Yang emphasized the importance of learning how the source of the content can be checked.

Follow the original source of the photo or video. Is it credible? When the video of a social media platform such as TikTok comes, read the label, hashtags and the description of the account profile for a mention of AI. Finally, check whether the message transmitted in the photo or video matches reports from reputable news agencies.