Thousands of fake videos are already circulating online which are often impossible to distinguish from authentic images. The issue is no longer technological; it has become societal. If we can no longer distinguish between what is real and what is fake, how can we continue to trust the information?
Do you remember this iconic video? It shows Will Smith eating a huge amount of spaghetti and becoming deformed as he swallows the pasta. Fascinating but grotesque, this video, which appeared at the dawn of generative AI in 2023, serves as a benchmark for measuring technological advances in AI image generation.
Since then, a threshold has been crossed. The symbolic turning point has a name: Sora 2. Trained on tens of millions of real videos, the model has learned light management, shadow consistency, and the physical continuity of human movements. The leap in realism is spectacular.
Sora 2 has also introduced a feature called « Cameos »: using a photo, a voice sample, and text instructions, it is now possible to integrate the face and voice of a real person into a fully generated scene. Initially, a watermark indicated that the video was synthetic. This was quickly circumvented.
The threat of deepfakes, these fake contents that are more real than life itself, is now a very tangible reality and within the reach of almost everyone. A viral example: the video entirely generated by AI and entitled « German Shepherd Saves Child from Eagle Attack, » which racked up more than 47 million views in two months!
Emotion supersedes verification
And this is just one example among many. The list of deepfakes that have marked our daily lives is growing longer by the day: the false announcement of war between Canada and the United States, the image of the Pentagon on fire that provoked an immediate reaction from the financial markets, the fake video involving Philippine President Ferdinand Marcos Jr., etc.
One of these deepfakes is even strongly suspected of having caused the defeat of Michal Šimečka, the unsuccessful candidate in the 2023 Slovak presidential elections. It was a fabricated audio recording, conveniently released 48 hours before the election, during the election moratorium, thus preventing any widespread denial. In it, the candidate was heard discussing with a journalist a plan to rig the elections, notably by buying the votes of the Roma minority. Some people still believe it…
This case proves that even when content is denied, it continues to circulate: emotion precedes verification.
Deceptive realism does not only affect information. It is also transforming human relationships. On certain forums, thousands of users post photos and videos of their AI-generated « partners. » The subreddit « My Boyfriend is AI, » which has over 40,000 members, is symptomatic of this new phenomenon. It brings together a small community where users exclusively share their experiences of romantic relationships with AI. Some even claim to truly believe in the existence of their virtual companion.
100% AI
Deepfakes can also have very serious impacts, including on our health. In Australia in 2024, the « Glyco Balance » scam used AI-generated videos showing real doctors promoting a dietary supplement for diabetes. However, the product is not listed on the official register of therapeutic products, and there is no credible scientific basis to support its effectiveness. Some patients nevertheless considered stopping their treatment, putting their lives at risk.
Another telling example, already mentioned in these columns: according to the music streaming app Deezer, 18% of the music published on the platform last year was generated by AI, or around 20,000 tracks per day. As a result, the majority of listeners cannot distinguish between human and synthetic productions.
Taken together, these phenomena go hand in hand with the emergence of 100% virtual stars, particularly influencers. Lil Miquela, launched in 2016, was one of the first virtual influencers. She has collaborated with Calvin Klein and Prada. In 2024, models such as Itana Lopez, generated entirely by AI, publish, monetize, and interact like real people. Some agencies justify this choice by the « ease of management »of these virtual personalities.
In 2025, the Particle 6 studio created the AI actress Tilly Norwood after testing more than 2,000 versions. Finally, a few days ago, the Chinese AI video generator Seedance 2.0 took the feat even further by generating entire sequences while maintaining consistency from shot to shot. In short, this tool from TikTok’s parent company can generate an entire film, surpassing known models. This is enough to give the creative industries a new scare and herald a wave of deepfakes.
Written in French by Zakaria Choukrallah; edited in English by AngloMedia Group.
