DIGITAL LIFE

Real or fake? AI videos simulate movie behind-the-scenes footage, confusing internet users and going viral on social media

In a matter of seconds, a video can travel across the globe, garner millions of views, and spark immediate reactions on social media. The problem is that, increasingly, these videos depict situations that never actually happened.

In recent years, content created by artificial intelligence has begun to replicate human faces, voices, and gestures with a level of realism capable of fooling even attentive users. In a fast-paced digital environment where verification often follows sharing, this type of technology has become fertile ground for misinformation.

A few famous examples illustrate the phenomenon. During the war between Russia and Ukraine, a manipulated video showed Ukrainian President Volodymyr Zelenskyy seemingly calling on soldiers to surrender. The content was fake but circulated rapidly online before being debunked. In another instance, a deepfake of Nvidia CEO Jensen Huang was used in a fraudulent broadcast to promote a cryptocurrency scam.

Amidst geopolitical tensions—such as the recent escalation involving Iran, Israel, and the United States—experts warn that manipulated videos could emerge to sway public opinion or fuel political narratives.

According to a study published by DeepStrike, the number of deepfakes available online grew from around 500,000 in 2023 to a projected figure of over 8 million by 2025, while fraud involving this technology surged by 3,000% during the same period. The study also reveals that only 24.5% of people can correctly identify fake videos when they are of high quality.

For creative director and digital strategist Náthan Ximenes, founder of NTX Group, the evolution of technology demands a new mindset from the public regarding the images circulating online. "For a long time, we believed that video was proof. With artificial intelligence, that has changed. Today, images can be easily produced and circulated before any verification takes place."

Given this landscape, learning to recognize potential signs of manipulation has become an essential skill for anyone consuming information on social media. Although artificial intelligence systems are becoming increasingly sophisticated, Ximenes points out that there are still technical indicators that can help identify artificially generated videos.

One of the most common signs lies in the synchronization between voice and mouth movements. In many deepfakes, the lips do not perfectly match the words or exhibit slight delays. Another area to watch is the eyes and facial expressions, which may appear stiff or repetitive, as replicating human micro-expressions remains a challenge for many video generation models.

It is also worth noting environmental details, such as subtle lighting changes on the face, slightly distorted hairlines, or shadows that do not match the setting. In some cases, the audio quality itself can raise suspicions. AI-generated voices tend to have a uniform intonation, lacking natural pauses, breathing, or emotional variation.

According to Náthan, the public needs to develop a more critical eye regarding digital content. "Technology has evolved very rapidly, making it possible to produce extremely convincing videos in just a few minutes. That is why checking the source and context is just as important as observing the video's technical details," he states.

Another essential precaution is verifying the content's origin. Videos shared by unknown profiles—lacking references to dates, locations, or reliable sources—warrant extra scrutiny. Often, a quick search for news reports or the official channels of the person mentioned is enough to confirm whether the event actually took place.

In a digital landscape where images can be easily fabricated, experts say that media literacy is becoming just as important as knowing how to use the technologies themselves. Now more than ever, seeing is no longer a guarantee of truth.

As Ximenes sums it up: "Artificial intelligence has opened up incredible possibilities for audiovisual production, but it has also brought a new challenge. Today, before believing a video, we need to learn to ask whether the event actually happened."

Videos supposedly showing behind-the-scenes footage of major film productions are going viral on social media, yet in many cases, the filming never actually took place.

One recent example features scenes of a massive water tank with artificial waves, film crews, and a set-piece ship. Posts claim this shows the making of *Tsunami* (2009), a South Korean blockbuster about a tsunami. In other versions, the same video appears as behind-the-scenes footage for different films—even *The Odyssey*, an epic by Christopher Nolan currently showing in theaters.

However, expert content creators point out several signs that the scenes are synthetic—such as Osmar Portilho, a journalist and videomaker with over 83,500 followers.

Clues include cameras that do not resemble real film equipment, disproportionate vehicles, people moving illogically on set, and an entire crew positioned next to a pool about to be flooded, with no protection for the equipment or the professionals.

Another detail stands out: the same city model appears in videos claiming to depict completely different locations, such as Paris and New York—an inconsistency typical of content produced by generative models.

The videos also exploit an element that tends to boost credibility: public curiosity about the behind-the-scenes aspects of major productions. By combining industrial settings, green screens, cranes, cameras, and special effects, AI tools can produce content that looks plausible at first glance.

mundophone