“China Musk” is just the tip of the iceberg of fake faces

Chinese netizens have always had a Silicon Valley hero complex, and they always like to "localize" Silicon Valley startup stars.

After Jobs, the most-discussed Silicon Valley entrepreneur star is now Musk, who is rich and rhetorical.

Those who admire Musk will say that China cannot produce a second dreamer who dares to speak and do, while those who hate Musk believe that China does not need another rhetorical capitalist.

But when "China Musk" exploded on Douyin, both those who admired and hated Musk fell silent.

This "Chinese Musk" who speaks a pure dialect and broken English looks so absurd and weird, after a long time you will find that he is somewhat reasonable.

Although the outfit and background of "Ma Yilong 0" in the video are not as refined as the former richest man in the world, the wildness in his eyebrows that doesn't take the world into consideration is so real.

When the audience saw him hold up a little blue bird made of paper and said "Love Tiger Oil", the frowning brows that had been frowning finally couldn't hold back. After laughing, they not only sighed that the hot awareness of earthy videos is so new.

Soon, the absurd meteorite slammed into the internet planet and went viral.

A good netizen posted the video on Twitter, asking Musk, who really fell in love with the little blue bird, what he thought of his "Chinese stand-in" on the other side of the ocean.

Obviously, the level of collision between the two makes Musk, who is used to seeing big scenes, unbelievable. Musk said that he is quite interested in meeting "Ma Yilong 0", provided that the collision Brothers really exist.

After hearing the interactive request initiated by Musk, "Ma Yilong 0" in Hebei also responded with a Weibo, saying that he also wanted to meet "Li Kui" himself.

Just when everyone was expecting a happy ending for this two-way story, the Weibo account of "Ma Yilong 0" was suddenly banned from Weibo this week, and many videos imitating Musk have been deleted.

When "Ma Yilong 0" exploded, many netizens already suspected that the video of "Ma Yilong 0" was made using Deepfakes face-changing technology. The real photos of the two show that the difference in appearance between the two is still very obvious.

After "Ma Yilong 0" was banned from Weibo, the face-hitting farce of "Li Gui meets Li Kui" has basically confirmed the fact that it is fake, as long as the Internet is given a little more time, this Hebei guy will soon be Forgotten in the long river of the Internet.

But after "Ma Yilong 0" disappears, who knows when the next "Ma Yilong 1" and "Ma Yinglong 2" will appear?

flooded fake face

"Imitating stars" has always been a tactic used by many unpopular amateurs to attract attention in a short period of time.

Click on short-platform live broadcast platforms such as Douyin and Kuaishou, and absurd scenes like "Zhoukou Jielun" PK "Junjie Shaobing" abound.

If it is said that the use of clothing modeling and makeup to imitate is the Stone Age of producing fake stars, then "counterfeiting" has already accelerated into the AI ​​era. The photos are used for machine learning to create a realistic effect like "Ma Yilong 0".

Last year, Tom Cruise made a splash in Tiktok in a similar fashion.

In the video posted by the account named DeepTomCruise, people found that Tom Cruise, who has always been handsome and cold, shared his daily life with everyone like a passionate big boy, and was unexpectedly talkative and humorous.

Combined with the Deep prefix of the username, it is clear that this is a post-production video synthesized with Deepfake technology.

But what amazes many people is that the completion of these videos is quite high, and it is difficult for ordinary people to see the flaws in them, so the video of "Fake Tom Cruise" went viral on Tiktok and quickly gained more than 1 billion views and 3.4 million followers.

The popularity of "Fake Tom Cruise" made its creator Chris Umé really feel the Internet's enthusiasm for "fake". He explained to fans how "Fake Tom Cruise" was made on his personal YouTube account. The core technology used is the well-known open source face swap project DeepFaceLab.

DeepFaceLab can be said to be the most popular way of AI face changing. Its author Ivan Perov confidently wrote in the project introduction on Github that "95% of Deepfake videos are currently made with DeepFaceLab".

The logic of DeepFaceLab's face-changing is not complicated. It can be roughly divided into several steps, such as detecting faces, adjusting angles, and replacing faces. In theory, as long as the algorithm is given enough face information to learn, DeepFaceLab can make fake face swaps. Effect.

In the production process of "Fake Tom Cruise", Chris Umé used a machine to learn a lot of Tom Cruise's real face, and found the "real face" with the most suitable angle to replace it.

Of course, the actual operation of AI face replacement is much more complicated than in theory. If you just replace the face, it is easy to make flaws when the character moves quickly, the face is blocked, or when you look in the mirror, so it is necessary to make a realistic The fake video still needs a lot of post-work.

Chris Umé said that in the early days of making the video, it took 2 months to train the AI ​​artificial model. After the face was replaced, he had to adjust it frame by frame to achieve the best results.

Now, after multiple iterations of "Fake Tom Cruise", ordinary face occlusions such as eating lollipops, wearing sunglasses, and drinking have basically no effect on fake faces. In the latest video, this handsome fake face can be "truthfully" restored even on the mirror, and the level of fidelity has reached a new level.

Chris Umé mentioned in an interview with The Verge that he believes deepfake technology will change the way people think about video editing in the same way that Photoshop changed photo editing in the past.

In fact, large companies including Snapchat, Disney, and ByteDance have carried out research related to DeepFaceLab technology and published a series of patents and papers.

▲ Picture from: Vice

But among the numerous face-changing videos, creators like Chris Umé who use it non-maliciously are still a minority, and more people use Deepfake as a tool to release evil, to create pornographic content and maliciously spread misinformation.

In a recent Vice report on deepfake technology, it was mentioned that 96% of deepfake videos in 2019 were pornographic, and dozens of works were uploaded every day on deepfake porn sites, and the vast majority of them were replaced The "face master" of Shangren's face has no idea about this.

The Deepfake video market is like a huge iceberg. The face-changing videos that people usually see are just the tip of the iceberg, and more invisible people's content is buried under the dark sea.

The proliferation of pornographic and malicious deepfake videos is undoubtedly a violation of the portrait rights of the replaced. Social platforms including Twitter have strict measures against malicious deepfakes, but it is still difficult to prevent the corresponding content from being broadcast on Telegram, Discord, etc. Spread on the platform.

Despite this, people's attitude towards Deepfake is still ambiguous. On the one hand, people hate malicious infringement of portraits and fraud, and on the other hand, people are eager to see more interesting face-changing works.

Deepfake is like a knife that everyone can afford. Some people use it to perform knife-swallowing tricks, and some people use it to plan terrorist attacks. Video tool.

"Fake face" is by your side

Just last week, my colleague bought a pair of original earphones that have already become popular. Due to its extremely low price and unexpected sound quality performance, there is a strong sense of contrast, many users who have worn it. It is joked that wearing it will trigger "tears of remorse".

Soon, Xinyan released a listening video. The video didn't say a word, but it showed the shock, surprise, and regret for not buying it earlier.

Just when I was amazed at his lens performance, I said that this was actually shot with Snapchat's new "AR Crying Face" filter, which is an out-and-out fake video.

"AR Crying Face" is arguably one of Snapchat's hottest filters right now, with more than 180 million users using it to take 1.3 billion shots in less than a month.

The popularity of "AR Crying Face" is largely due to its realistic effect. As long as you turn on the filter and take a pat on your face, you who were laughing in one second will be crying the next second, and the more you laugh, the more you laugh. Bright "crying" is even more miserable.

This is actually using Deepfake technology to analyze and replace faces in real time. Although it is not as refined as "Fake Tom Cruise", it is more than enough to tease friends.

After the "crying face" exploded, some netizens suspected that Snapchat found inspiration on Depp's ex-wife Heard, and then made the same crying face.

Snapchat then explained that "crying face" was actually planned and developed 6 months ago, and a "smiley face" filter was developed together with it, but after some measurement, Snapchat thought that crying face would be more popular, so it was abandoned. smiley.

It turns out that people are indeed obsessed with the way they cry to the point of collapse, but the question arises again, is it you who has been replaced with a fake face?

Similarly, if Musk's face is put on "Ma Yilong 0", what is true and what is false? Where does people's perception of "true identity" come from?

This is the "Theseus Paradox" after people's lives are invaded by Deepfake technology.

Last year, the famous game anchor Wuhu Da Sima was made by fans to make a face-changing video called "Muscle Golden Wheel", which received more than 10 million views at station B, causing a wave of face-changing craze.

The "Muscle Golden Wheel" series of videos changed the face of Wuhu Da Sima to a well-built fitness blogger. The familiar and unfamiliar images were superimposed together, forming a strong visual impact, making the video in the game. The circle quickly became popular.

Interestingly, after the explosion of "Muscle Golden Wheel", the most discussed in Wuhu Da Sima's live broadcast room is no longer the game, but urging the anchor to exercise to become a real "Muscle Golden Wheel".

It seems that the virtual film synthesized in their cognition is real and perfect, while the person with the real face in front of them is full of flaws.

Obviously, people are not ready to live in harmony with the "fake face".

In addition to causing confusion to people's identity cognition, "fake face" also has potential social risks. During political elections or wars, deepfake videos often become a manipulation tool for some people with intentions, using real faces familiar to the public to spread false news to achieve the purpose of shaping public opinion and inciting conflict.

Therefore, since we have sharpened the "spear" that can destroy the online world, we should also build a "shield" that can withstand its attacks.

However, compared with the popularity of Deepfake, there are far fewer anti-Deepfake or deepfake screening actions on the Internet.

The Massachusetts Institute of Technology announced a project called Detect DeepFakes last year, which aims to help improve people's ability to identify Deepfake videos on a large scale in the future by training detection algorithms, but after a year, we still can't see it. to more substantial tools to help people identify synthetic videos.

Nowadays, people's virtual life is more and more intertwined with real life, and the authenticity of the virtual world has become blurred. If we cannot have an appropriate technical means to define "false", then "true" Where did it come from?

Stop talking nonsense.

#Welcome to pay attention to the official WeChat account of Aifaner: Aifaner (WeChat: ifanr), more exciting content will be brought to you as soon as possible.

Love Faner | Original link · View comments · Sina Weibo