TOKYO - From a single photograph, AI can now generate a movie-like video in as little as four hours, raising urgent questions about what precautions society should take and how such content can be detected.
A video posted on social media appears realistic at first glance but is in fact entirely fake, with close inspection revealing that while opposition leader Noda’s head moves, the shadow behind him remains fixed, along with other unnatural elements such as another party leader’s head overlapping with on-screen captions.
According to Daisuke Furuta, editor-in-chief of the Japan Fact-Check Center, the number of manipulated and fabricated images and videos created using generative AI has increased sharply since 2025, posing a growing threat ahead of elections.
So how are these AI-generated videos made.
In one example, former U.S. President Donald Trump is shown singing about world peace, followed by Russian President Vladimir Putin passionately singing about the value of life, with both videos entirely created using generative AI.
The videos were produced by AI video creator Taiga Aoyama, who creates and distributes AI-generated videos themed around well-known politicians on YouTube.
“Right now, if you study a little, anyone can create fake images,” Aoyama said, adding that this is simply the reality of current technology.
Aoyama began distributing videos in July 2025, saying he turned his attention to political figures after seeing online excitement surrounding the House of Councillors election.
“Politicians have a certain level of publicity rights,” he said. “I think it’s wrong to demean them, but if the content helps explain their views or positions in a clearer way, that kind of use could be acceptable.”
Aoyama says he taught himself how to create AI videos by researching online, and demonstrated the process using a single photograph of reporter Yota Muroya, taken during an interview, explaining that just one image is enough to generate a video.
He enters a prompt instructing the AI to create images resembling a samurai-style movie scene, as if the subject were fighting in the Sengoku period, then feeds the original photograph into the system, producing a new image in about 30 seconds.
After repeating the process until the desired image is created, he then instructs the AI to convert each image into individual video cuts, with music also generated entirely by AI, completing the process in roughly four hours.
While Aoyama says he carefully researches political figures in advance and deliberately produces only clearly fake videos to avoid misleading viewers, he also acknowledges the dangers posed by AI-generated content.
“If someone without ethics studies a little and decides to spread fake news deliberately, they can distribute as much misinformation as they want,” he said. “Viewers really need to be cautious.”
As fake videos proliferate, researchers are also developing AI systems to detect them.
In one demonstration, two videos are shown side by side, one real and the other created using generative AI, with differences difficult to spot at first glance.
Professor Isao Echizen of the National Institute of Informatics explained that AI-generated faces can now be automatically flagged with a red frame, allowing systems to identify whether a face is real or AI-generated.
This is achieved by training AI on large volumes of both real images and AI-generated images, enabling it to distinguish between the two.
As AI creates fake videos and AI detects them, Echizen cautions that human judgment remains essential.
“It is extremely important for users themselves to verify information by consulting reliable sources such as newspapers, television, and the results of fact-checking organizations, rather than relying solely on social media,” he said.
Source: TBS
