The Double-Edged Sword of Face-Swapping Technology: From Hollywood Magic to AI Frauds

Thanks to the popularity of Wonder Woman, Gal Gadot has become the ideal woman for thousands of men. But as the number of searches skyrocketed, some “careful” netizens discovered something remarkable: a pornographic movie starring Gal Gadot?

The collision of “celebrity” and “mini-movie” is big news everywhere, and if Wonder Woman had really acted in a mini-movie, she couldn’t have done it so quietly. There’s only one answer: it’s a fake.

This video, at first glance, will really make people think that “Wonder Woman” Gadot acted. But on closer inspection, there are still some flaws: the contours of the face are not smooth, and the skin color at the edges has undergone some abrupt changes, as if wearing a thin mask.

“Of course, Wonder Woman didn’t go to sea! It was a Reddit user who “photoshopped” Gal Gadot’s face into the face of the actress in an adult video. The video has now been taken down from the site. Gadot isn’t the only actress who has been “shot”.

How did the Reddit user make the video? The key to successful face-swapping is to synchronize the face with the original video, and he used machine learning to achieve this effect.

In fact, video face swapping is nothing new; the technique has long been widely used in blockbuster movies and games.

**Magic Face **Transplant

Face-swapping technology in blockbuster movies has become more sophisticated. Remember Davy Jones, the disgustingly tentacled captain of Pirates of the Caribbean? He was portrayed using face-swapping technology, as were the Orcs and Gollum in The Lord of the Rings, Smaug in The Hobbit, and the protagonists of Avatar, to name a few.

The vast majority of commercial productions have simply used this technique to create fantasy creatures, however Industrial Light & Magic has taken it a step further and used face-swapping to bring deceased actors back to life on screen.

Horror actor Peter Cushing, who died in 1994, was brought to the modern day giant screen by technicians in the 2016 movie Star Wars: Rogue One. Behind the “modern” actor was another actor, Guy Henry, whose expression was captured by technicians who then composited a digital version of Peter Cushing. Although the faces of the two actors are quite similar, the technicians still had to deal with a lot of details, such as the fact that Peter Cushing does not move his upper lip when he pronounces the sound “ah”.

This technique sparked a lot of controversy, with some people saying it was unacceptable. Although the movie has been granted a portrait license, there are still some ethical questions that need to be answered.

Face-swapping has not only been seen in movies, it has also been used in games. Hideo Kojima’s Death Stranding is an open-world action game, and actor Norman Reedus, who is famous for The Walking Dead, will “star” in the game, and his image will appear in the game to promote the development of the game’s plot.

A few days ago, the Hong Kong police disclosed that a multinational company’s Hong Kong branch of the financial staff, in order to participate in an internal “multi-person video conference”, was cheated 200 million Hong Kong dollars. It turned out that, in addition to the victim is “real”, the video conference in the other “colleagues” are using AI face-swapping technology impersonation.

This year’s news has also exposed a number of cases of fraud using AI voice mimicry and face-swapping. Behind these frauds are deep forgery technology Deepfake’s figure. This is not a new technology, but the outbreak of generative AI technology in the past year has greatly lowered its threshold, making it easier and easier to produce unrecognizable face-swapping videos, “there is a picture of the truth” has become no longer reliable, and the scam is not defensible. So how do we protect our personal information? And how to recognize whether the dialog video is a “real person” or a “digital person”?

How exactly is face-swapping accomplished?

To put it simply, the general principle of face changing technology is to recognize the specific structure of the face through algorithms - not only the position of the eyes, nose and mouth, but also the shape of the cheekbones, chin and cheeks, and then do the replacement according to these characteristic points.

Strictly speaking, this technology belongs to a small category of motion capture technology, called face capture, and we often see the yellow dots on the face of the motion-capture actors, is the key point in the change of expression, like the face of the “joints”, the computer only needs this information to synthesize the expression. In facial recognition technology, it is called face tracking.

The key to motion capture technology is to recognize human expressions, especially those key points on the face. And deep learning technology can fulfill this requirement.

Deep learning is a branch of machine learning. One of the better deep learning algorithms for facial capture is Face2Face, which is capable of real-time face capture and synchronized display of RGB video. For example, a researcher wants to make the Trump in the video make the same expression as himself, what should he do? Here’s how to do it:

Step 1: Obtain a video sequence of Trump, called the target video;

Step 2: Obtain a video of the researcher’s expression using a camera device, called the source video;

Step 3: Track the source video to the target video and perform a real-time deformation transfer between the two by means of a dense photometric consistency measure.

Of course, deep learning gets much rougher results and is easier to recognize compared to those specialized film and video industry techniques, however it is also much cheaper. The software technologies and platforms of big companies like Google, Microsoft, and Amazon are publicly available, deep learning technology is growing by leaps and bounds, many papers are shared among researchers, and deep learning algorithms can be handled by home-grade graphics cards without necessarily using technologies like cloud computing and supercomputers.

Face swapping, once a complex and time-consuming process reserved for Hollywood special effects teams, has been democratized by Faceswap.tech. It’s no longer a domain exclusive to big-budget productions; instead, it’s a tool accessible to anyone with an internet connection and a desire to explore the fun and fantastical world of visual manipulation.

At its core, Faceswap.tech leverages deep learning algorithms that have been meticulously trained to recognize facial features, contours, and expressions with remarkable accuracy. This allows the platform to map one face onto another with unprecedented precision, creating seamless transitions that often defy the eye’s ability to discern the original from the swapped image.

2023 ChatGPT suddenly burst fire brush screen, generative AI “fly into the common people’s home”, especially the introduction of video generation model Sora, greatly reducing the threshold of AI face replacement technology.OpenAI CEO Altman recently said at an event, “” OpenAI can reduce the cost of very high-quality AI technology to close to zero”.

OpenAI can reduce the cost of very high-quality AI technology to close to zero”. But the use of technology is a double-edged sword, over the past year, the use of Deepfake technology to forge false images, video and audio to impersonate others to carry out fraud more and more incidents, “AI fraud success rate of close to 100%”, “AI face-switching cheated a Hong Kong man 200 million Hong Kong dollars ”, “company boss by AI fraud 4.3 million” and other related topics frequently on the hot search, to the public property, privacy and other rights and interests.

How to recognize the “AI face”?

Some experts say that you can have the person press their nose and face and observe their facial changes. If it is a real person, the face will be deformed, if it is a digital person, it will not, or appear broken.

However, some industry insiders believe that this method does not always work. Zhu Kai, an expert in security algorithms at Ant Group’s Tiangui Laboratory, said in an interview with Jiefang Daily that this approach has some reference value, but it may not work completely.

Because most of the generation algorithms (AIGC) to generate a clean face, and press the nose or press the face, means that the face to add shade, the relevant algorithms may be due to the training of insufficient data, and thus in the generation of obscured face breakage.

However, some more advanced algorithms can already generate images in static scenes with partial occlusion. In other words, some algorithms are able to give a fake effect if you just press the nose or face.

As of now, Zhu Kai suggests shaking the head from side to side while holding down the nose or face, so as to recognize AI face-swapping in a moving way. Because many of the current AI face swap is based on the photo of the person being imitated, the photo is flat, and the generated “digital person” is also mainly frontal, flat.

When the face moves, it will produce a crack. In addition, China Computer Federation Security Professional Committee of the Digital Economy and Security Working Group member Fang Yu said in an interview with CCTV, you can ask the other party in the video call, in front of the face by waving to recognize.

This approach can cause interference with facial data, and faked videos can show abnormalities such as shaking or flashing. In peer-to-peer communication, on the other hand, it is possible to ask questions that only the other party knows to verify the authenticity of the other party.

It is worth mentioning that Deepfake attack and defense is a relative and constantly refined process, you are progressing, Deepfake is also progressing, “what we have to do is to run in front of it”, said Tao Ye, product director of ZOLOZ. AI against AI, also known as “magic to defeat magic”, is actively practiced in the industry.

Baidu, Ali, Tencent, ByteDance, Ant Group, and Tech Data Xunfei and other AI head enterprises are actively improving their ability to recognize in-depth forged content by developing advanced AI algorithms and tools.

For example, Ali develops a variety of algorithms by analyzing the subtle features of videos, while ByteDance and Tech Data Xunfei combine content analysis and speech recognition technologies to maintain the authenticity of cyberspace.

In addition, ZOLOZ, a technology brand under Ant Digital Technology, recently released ZOLOZ Deeper, a comprehensive forgery prevention and control product, to intercept the risk of “AI face-switching” in the process of user face-swiping, which has been firstly applied in the field of identity security.

According to reports, ZOLOZ launched the Deepfake algorithm research and development as early as 2019, and at that time, it mainly dealt with some simple face attacks, such as pictures, videos, masks and other false content presented in front of the camera to break the face recognition system.

ZOLOZ Deeper is for injection, activation and other complex attacks iterative product, in the service of Indonesia, a head of the bank, the product on-line within a month to achieve the Deepfake risk “0 missed”.

Tao Ye disclosed that ZOLOZ and Ant Security Response Center set up a multi-million dollar bonus pool to support security geeks to dig ZOLOZ Deeper vulnerabilities, and submit vulnerability information through the official website of “Ant Group Security Response Center”. “There is no absolute security in the Internet world, and we hope to unite external security technology talents to continuously improve ZOLOZ’s security level and guard users’ sense of security.”

Conclusion

Many people have expressed great concern about the use of “AI face-swapping” for pornographic videos. Some netizens said, if my ex-boyfriend or ex-husband, I still hold a grudge against me, it is possible to use this less complex technology, my head will be installed in the pornographic film heroine, and then in the social media to disseminate, so as to humiliate me how to do?

Other netizens say that this technology is still just developing, and as the technology becomes more sophisticated in the future, will it harm more women? Others say that technology is not good or evil, but has boundaries and must be used within the law. If it is only for personal study, there is no problem at all.

Faceswap.tech represents a significant milestone in the evolution of visual entertainment and creative expression. By harnessing the power of AI and computer vision, it has transformed face swapping into an accessible and enjoyable activity for millions of users worldwide.

As the platform continues to evolve and expand its capabilities, it promises to open up even more avenues for creativity and innovation. Whether you’re a casual user looking for a fun way to pass the time or a professional seeking to enhance your content creation workflow, Faceswap.tech has something to offer.

In conclusion, Faceswap.tech is more than just a face-swapping tool; it’s a testament to the limitless potential of technology when it’s harnessed for the purpose of entertainment, education, and connection. As we embrace this exciting new era of visual entertainment, Faceswap.tech stands ready to lead the way, inspiring and delighting users around the globe.