Seeing is not believing.AI face-changingDeceived many people.
Two days ago, Hubei police cracked a case of using AI face-changing technology to unlock social accounts and resell them for profit. The criminals used AI face-changing technology to help people unlock dating accounts that were frozen by the platform, charging 15 to 30 yuan for each unlock. The gang unfrozen a total of 2,107 accounts and made a profit of 400,000 yuan.
Data shows that AI-based deepfake fraud surged by 3000% in 2023. Deepfake is a combination of learning and fake, which mainly refers to the use of AI deep learning technology to synthesize a person's picture, video, or even voice. The most common is AI face-changing. The Ministry of Industry and Information Technology recently stated that it will launch an AI face-changing fraud risk reminder function.
On the one hand, AI face-changing fraud cases occur frequently, while on the other hand, AI face-changing technology is constantly iterating.
The AI community has recently been in an uproar over an AI face-changing project called Deep-Live-Cam. Users only need to upload a photo to instantly change their face into Zuckerberg, Dong Yuhui, or anyone else they want to be.
This project is free and open to all users. It is published on GitHub (code hosting platform) and supports installation on personal computers. The project developers have announced the installation method, so ordinary people without technical experience can learn while installing.
Due to its simple and convenient operation and high degree of face-changing authenticity, Deep-Live-Cam topped the GitHub hot list as soon as it was released, and received more than 1,600 stars in 24 hours (stars represent project likes and attention indicators, and thousands of stars are a very good level). Many netizens changed their faces to look like Musk and experienced the feeling of being a global billionaire.
But while everyone is impressed by the breakthroughs in AI face-changing technology, they are also worried that "fraud has become easier." Should we resist AI face-changing?
Ordinary people can use AI face-changing, the cost is as low as a dozen yuan
There are many AI face-changing tools available at home and abroad. The reason why Deep-Live-Cam has become a hot topic is mainly because users only need to upload a picture but the replacement effect is very realistic, and it also supports real-time face changing.
In fact, the technical threshold of the "real-time face-changing" function is not high.
Xiao Zihao, co-founder and algorithm scientist of Ruilai Intelligence, said that there are two ways to achieve the effect of real-time video face-changing: one is to play a pre-made face-changing video, and the other is to access the real-time face-changing video. Deep-Live-Cam belongs to the latter.
Regardless of the type, AI face-changing videos must take into account the continuity of the previous and subsequent frames. Based on the processing of continuous face videos, slight adjustments can be made based on many detection resources of the previous frame. "This also makes video/live face-changing based on continuous frames more efficient than single face-changing based on independent frames," said a researcher engaged in AI technology.
The underlying technology that AI face-changing relies on is Deepfake. Practitioners speculate that Deep-Live-Cam also uses this technology, so judging from the materials and computing power spent on production and synthesis alone, the cost will not be very high.
Xiao Zihao said that if other hardware purchase costs are not taken into account, most AI face-changing tools cost only a dozen yuan each time, mainly including photos and computing power, and Deep-Live-Cam may be even lower. Compared with other AI face-changing tools, Deep-Live-Cam uses fewer photos and can also be used on ordinary computers, so the cost of photos and computing power is lower.
In contrast, Deep-Live-Cam supports face-changing with only one picture, which is even more shocking to practitioners. This is because previous AI face-changing products required users to provide at least a dozen photos from different angles.
The reduced difficulty of taking photos has made users excited. AI follower Xiao C said that he had never used any AI face-changing tools before. He felt that Deep-Live-Cam had a low operating threshold and could be completed by following the publisher's instructions. It was not difficult to find face-changing materials either.
He changed his face into a group of celebrities such as Guo Degang, Wu Jing, Lei Jiayin, etc., and only used one picture of each of them. He said that as long as the photo provided is of the front face, with no obstruction on the forehead, natural light, close to one's own skin color and ID photo proportions, one photo will be successful, and the face-changing effect will be very natural.
More than one practitioner believes that the AI face-changing technology used in some TV dramas is not only difficult to access and expensive, but also not natural enough, with deformation and distortion often occurring during speech. "Resemblance is the biggest problem with AI face-changing tools. Deep-Live-Cam can support a picture to be broadcast live without obvious traces of stickers on the edge of the face, and the skin color is not much different from the original character. The realism is ahead of many AI face-changing tools," Xiao Zihao analyzed. Its developers should have achieved a breakthrough in the image rendering model that previous AI face-changing tools did not have.
However, Deep-Live-Cam still has some bugs, such as supporting face-changing but not voice-changing. But Xiao Zihao said that it is not difficult to achieve this, just connect to the real-time voice-changing tool, and anyone with a little programming knowledge can do it.
AI face-changing: fraud and infringement that cannot be avoided
Tools such as Deep-Live-Cam have lowered the threshold for users to use them, but people are also beginning to fear that criminals can use just one picture to change faces and commit fraud. For ordinary people, the risks will increase dramatically.
Ruilai Smart Data pointed out that the amount involved in domestic AI fraud cases increased from 2,000 in 2020 to 16.7 million in 2023, with a compound annual growth rate of 1928.8%. In the first half of 2024, AI fraud cases exceeded 185 million. The "2024 Artificial Intelligence Security Report" released by Qi'anxin showed that AI-based deep fake fraud increased by 30,00% in 2023.
AI voice-changing, AI face-changing, and AI generation are three common methods of AI fraud. Among them, AI-generated content has limited seconds, and the image quality and character flaws are obvious, so criminals most often use AI voice-changing and AI face-changing. Most people believe that "seeing is believing", which makes AI face-changing more deceptive than AI voice-changing.
AI face-changing not only deceived the elderly, but also caused many young people working in the technology field to lose millions overnight.
According to public media reports, last year a CEO of a technology company received a request for help from a friend who transferred money, but was defrauded of 4.3 million yuan in less than 10 minutes. The scammer mainly used AI face-changing technology. Another employee of a multinational company in Hong Kong was deceived into a senior executive video conference designed by criminals, and transferred 200 million Hong Kong dollars to the other party's account.
It is a consensus among practitioners that AI face-changing has reached the level of being “indistinguishable from the real thing”.
Li Xu, the head of the anti-fraud team, is interested in AI face-changing technology because of various fraud cases. He said that AI face-changing fraud cases have occurred frequently in the past two years, and there are about three AI face-changing fraud methods on the market that are very mature.
The first is to use AI to change faces and forge pornographic videos to blackmail victims.
For example, criminal gangs will collect pictures and video information of victims from various channels such as the Internet, and then perform synthetic processing such as one-click stripping and AI face-changing, and send forged indecent videos or nude photos to the victims for blackmail.
The second type is using AI to change faces and impersonate acquaintances to commit fraud.
For example, fraud gangs will use some face-changing tools on the market to disguise themselves as relatives, friends, colleagues, leaders, etc. of the victims. They will first fabricate an emergency and use the urgency of the target to commit fraud. The victims are often prone to relax their vigilance and transfer the money out in a hurry.
Now, using Deep-Live-Cam, you only need to collect one or two photos of the victim to deeply imitate a video. Li Xu feels that the extent of AI face-changing fraud is beyond the cognition of most ordinary people, and has directly changed the script of telecommunications fraud, which is difficult to guard against and is getting worse and worse.
Third, they use AI to change faces and impersonate celebrities to conduct live streaming and sell goods, thereby deceiving users into placing orders.
For example, the anchor uses "AI face-changing" to "replace" his own face with a celebrity's face in the live broadcast room, thereby increasing the live video traffic and helping live broadcast sales.
In addition, infringement issues such as creating long and short videos of celebrities/ordinary people and making emoticons through AI face-changing without their permission are also very common.
AI face-changing fraud cases occur frequently and use various methods, which is closely related to the maturity and breadth of this technology.
Hanhao Technology CEO Wu Jieqian told "Focus", "At present, AI video generation, digital humans and AI face-changing technologies are all very mature, and AI face-changing is the earliest developed field among them."
The emergence of Deepfake technology in 2014 marked the birth of AI face-changing, but it has been spread in a small range until 2017, when some users used the technology to post false pornographic pictures. AI face-changing then came into the public eye, but at the same time, the technology has been controversial because of the large amount of fake content it has brought.
More than one practitioner said that in recent years, although the public discussion on AI face-changing has focused on saving film and television dramas, due to the difficulty in supervision, the use of AI face-changing for fraud and fraud has never stopped. In addition, AI face-changing technology has been iterating, and there are many players. At present, in addition to some large companies involved in this business, many domestic digital human and AI-generated video startups have developed AI face-changing businesses. The technology is relatively popular, and there are more and more such tools.
The technology is mature and widely used, and supervision is somewhat difficult, resulting in the abuse of AI face-changing technology over the years.
How to identify increasingly realistic AI face-changing?
In the past two years, there have been many illegal and criminal acts caused by AI face-changing. The uncontrollability has led some people to think that they should resist AI face-changing technology.
However, many practitioners believe that technology itself is neutral, and the good or bad impact it brings mainly depends on the user. The emergence and iteration of AI face-changing technology should not be stopped.
Wu Jieqian said that at present, there is no one-size-fits-all approach to AI face-changing technology both at home and abroad, and more legislative management is being carried out.
For example, the United States has enacted special legislation against deep fakes, the European Union has incorporated deep fakes into existing legal frameworks such as the General Data Protection Regulation, and countries such as Germany, Singapore, and the United Kingdom have also included deep fake-related crimes into their criminal law systems. China has also clarified the regulations on this technology in the Regulations on the Management of Deep Synthesis of Internet Information Services.
Many practitioners believe that it is necessary to introduce management measures such as labeling content created by AI and to teach the public some methods to prevent AI fraud.
For example, a technician once suggested that during a video call, one could ask the other party to put their fingers on their face and carefully observe whether the face is deformed, or ask the other party to open their mouth to see whether the teeth and tongue are clear and complete, so as to determine whether the image is a real person or an AI synthesis.
This method was quite effective last year, but with the upgrade of AI face-changing technology this year, it has become ineffective.
Xiao Zihao said that the current AI face-changing technology can achieve the goal of not deforming the face and movements when obstacles appear, and the image quality is very high. Li Xu also found that the recent AI face-changing fraud cases all used very powerful technical means, and it is almost impossible to distinguish them simply by movements.
In view of the rapid development of AI face-changing technology, Li Xu believes that being vigilant is a relatively effective method at present. For example, when receiving a video call from an acquaintance asking for money, you can ask the other party some private questions that only you know, and ask more people around you to verify.
Some practitioners think that using AI to identify AI may be a good idea.
For example, first check whether there are some elements that obviously violate common sense in the picture, focus on whether there are flaws on the edges of the face, and then use some AI tools. Xiao Zihao believes that using AI to identify AI can identify most scams. Currently, many companies and government departments have asked them to cooperate on the AI face-changing defense system.
However, this move has also caused some controversy at home and abroad.
First of all, this method has a certain usage threshold, and users need to download relevant applications on their mobile phones/computers.
Secondly, in essence, this type of defense tool has a certain lag, and is mainly used to counter current AI face-changing tools, but AI face-changing technology and synthesis methods are constantly changing.
Many people hope that relevant laws and regulations will be improved, such as increasing the penalties for illegal activities such as AI face-changing fraud, selling AI face-changing pornographic videos, and using celebrities with face-changing products to promote products.
Last year, the People's Court of Xiaoshan District, Hangzhou, sentenced a case of using AI face-changing to produce obscene videos. The defendant provided customized AI face-changing services, produced a large number of obscene videos through AI face-changing, and used the method of setting up groups to attract viewers to pay to join the group. The total amount involved in the case was more than 60,000 yuan. He was eventually sentenced to seven years and three months in prison and a fine of 60,000 yuan.
However, I would like to remind everyone that not all AI face-changing will infringe on portrait rights.
In June this year, the Beijing Internet Court pronounced a verdict on two AI face-changing infringement cases. Two Chinese-style short video models found that their videos appeared in a face-changing APP on the social platform, so they took the technology company behind the APP to court, claiming that it had infringed their portrait rights and personal rights.
The court's ruling was that the defendant was found to have used the video in which the plaintiff appeared, but because the plaintiff's face in the video in question had been replaced, the plaintiff's portrait rights were not infringed, and therefore it was only found that this action constituted an infringement of the plaintiff's personal information rights.
At this moment, AI face-changing technology is still being upgraded and has not yet been banned, and AI face-changing tools are still circulating in the market in large quantities.
Recently, when searching for "AI face-changing" directly on an e-commerce platform, no information can be found. However, as long as the keywords are slightly changed, relevant face-changing tools can still be found for sale, and the path is more secretive, jumping to a second-hand website for trading. At a time when technology and scams are constantly upgrading, what everyone can do is to be vigilant and prevent being harmed by AI face-changing.