A.I. Can Now Create Lifelike Videos. Can You Tell What’s Real?

Oops, not quite. This was made with A.I. tools. This is a fake fashion show generated by the appropriately-named Runway, an A.I. company. It was made by the Dor Brothers, a Berlin-based A.I. visuals company run by Yonatan Dor. The company, which has 12 employees, typically starts with A.I.-generated images, which are tweaked before using an image-to-video A.I. tool to bring them to life. This video, however, was made using a separate text-to-video tool from Runway.

“It’s better to make really high quality, detailed images and then animate them, instead of going directly from text to video,” Mr. Dor said in an interview.

Oops, not quite. This was manipulated using A.I. tools. It is not a real video game. It was created by Tim Yao, a software engineer based in Australia who produces A.I. artwork under the name Blind Sauce. He used a text-to-video tool by Kling AI, a tool based in China.

“I have to imagine myself as a film director, trying to guide the A.I. using only language to create the video,” he said in an email. “At times, it’s challenging to convey exactly what I want because certain nuances or details are hard to express purely through words.”

Parts of the video look like a realistic first-person shooter video game, but the video begins to show cracks as it continues, such as when the monster appears to disappear behind a ball of flames.

Oops, not quite. This was not made with A.I. tools. This is an ad produced by Nike and published on YouTube Shorts, a vertical video version of YouTube. A.I. generators excel at creating otherworldly elements and unlikely clashes, like a basketball player amid an orchestra. But the same systems may struggle with the complex physics and camera movements in this clip.

Oops, not quite. This was not made with A.I. tools. This is a genuine demolition of abandoned condominium buildings in China. Some A.I. generators tend to repeat similar elements, like faces or buildings, as they work to complete an image. The multiple identical buildings here could give an appearance of being generated by A.I. even though it is real.

Oops, not quite. This was made with A.I. tools. In what may be a world first, Toys ‘R’ Us created this commercial from A.I. clips. They used Sora, OpenAI’s video generator, to create the ad. It tells the story of Charles Lazarus, the founder of Toys ‘R’ Us.

OpenAI, the creator of ChatGPT, developed the sophisticated tool but said it had released it only to experts so far to “assess critical areas for harms or risks.” Meanwhile, many other A.I. companies have released cheap and easy-to-use A.I. video generators to the public. Here is the full ad:

Oops, not quite. This was manipulated using A.I. tools. The video is partially real — it includes a video originally filmed by Ollie Muhl, a TikTok influencer with more than 3.5 million followers. But the face has been swapped with another person’s using an A.I. video generator.

“Social media platforms must establish ethical guidelines, ensure transparency about A.I. use, and develop methods to detect deepfakes,” Mr. Muhl said in an emailed statement. “Despite its dangers, A.I. face-swapping can be a powerful tool for creativity and innovation if used with caution and respect.”

In the video, there is a brief frame near the end where the face-swapping A.I. appears to fail. Here is how the two compare.

Oops, not quite. This was made with A.I. tools. It was created by Nicolas Neubert, a member of the creative staff at Runway, an A.I. company.

To make the video, he used an A.I. image generator to create a still photo of a keynote speaker. Then he input that into Runway’s image-to-video generator. He added a simple prompt, instructing the A.I. to create a video with “natural movement.” The entire process took about five minutes.

“If we look at making films or bigger projects, there’s a little bit more involved,” he said in an interview. “But for something like that, it was really just over the afternoon.”

Though the video is remarkably lifelike, there are some errors: the woman’s left hand and arm seem to blend into each other at times. Some A.I. systems still struggle to render details like hands correctly, but they are rapidly improving.

Oops, not quite. This was not made with A.I. tools. It shows a real helicopter, but the camera’s shutter is moving at the same speed as the helicopter blades, making it appear as though the blades are not moving.

Though A.I. generators can often confuse objects, it is actually difficult for current A.I. generators to create videos like this. A.I. tools are trained on mass amounts of data, including scores of videos showing helicopters flying through the air — and very few videos like this one. In tests by The New York Times, A.I. generators could not create a video where the propellers were still.

Oops, not quite. This was not made with A.I. tools. It was filmed by Marco Bassot, a mountain biker and snowboarder who uploads thrilling first-person videos to his Instagram account. It depicts a harrowing ride along a thin ridge of the Dolomite Mountains in Italy.

Video generators have been trained to create first-person videos similar to this one, but the results tend to be imperfect. The bike can appear to slide along the path rather than bounce with the terrain.

Oops, not quite. This was made with A.I. tools. It was created by Blaine Brown, an artist who creates A.I. artwork under the name Blizaine. Mr. Brown used a text-to-video feature from Kling, an A.I. company, to make the clip. He started by asking the A.I. to generate a video of a man riding a shark floating over the beach. Then he changed the prompt to a man riding a jet ski on the ocean. The A.I. was able to smoothly transition between the two concepts in the complete clip, seen below: