4.3 million yuan in 9 minutes! Artificial intelligence face changing technology, who can get the initiative?

The legal representative of a Fuzhou-based technology company was deceived within 9 minutes by a person who falsely claimed to have a face falsified through AI technology at a price of 4.3 million yuan. Meanwhile, in Anhui, Mr. He was defrauded of 2.45 million yuan in just 9 seconds by a fraudulent activity through face-changing technology.

In the live broadcast room, celebrity-faced anchors such as Yang Mi and Di Li Gerba often appear, making it difficult for the audience to distinguish their true identities. At the same time, these anchors also bring goods through the live broadcast room.

Some people question that this promotional image is the result of Yang Mi's endorsement produced by the face-changing technology, and the picture comes from the video screenshot.

In presentation videos on social networking sites, a one-second face swap video can be seen everywhere, and anyone can make a video using artificial intelligence, and scammers are no exception.

Before we know it, each of us has stepped into the cyber world. Artificial intelligence has become increasingly popular, such as music and painting produced by artificial intelligence, as well as applications that can change the face and voice, such as "one-click unlock" and "one-click makeup", and even the advertising of artificial intelligence Sun Yanzi has become a witness of artificial intelligence.

While everyone is immersed in a technological feast, another side of technology gradually emerges: counterfeiting technology and technical identification seem to be playing a game of "cat and mouse". Finally, who will have the initiative?

AI face changing technology can be mass produced, how low is its technical threshold?

Face-changing technology using artificial intelligence is nothing new. In 2019, the advent of the "ZAO" software caused a sensation, people changed faces through the software, and a large number of spoof videos appeared on online video platforms, and many celebrities were maliciously attacked and spread false information.

However, despite this, the "ZAO" application at the time was quickly removed due to privacy violations and other issues, but the negative effects of this technology have always existed. It was not until the popularization of artificial intelligence technology this year that face-changing applications rose again.

Xiao Zihao, co-founder and algorithm scientist of RealAI, pointed out that deep synthesis technology is mainly used in artificial intelligence face and voice changes. With the increase of open source and product services of deep synthesis technology, the threshold for deep synthesis production has gradually decreased, and the technology has been "civilized". The average user can create a deep synthesis effect using only a small amount of image, audio, and other sample data, and using easy-to-use compositing tools.

Xiao Zihao introduced that there are two ways to change the face of artificial intelligence in the current video chat or live broadcast: one is to play a pre-made face change video, and the other is a real-time face change video. The first option has very low production costs and is supported by mature open applications; While the second way does not expose the application, there is mature technical support. The time it takes to make a video is related to equipment, computing power, and other factors. Current technology can complete the output in the same amount of time as generating video. Making these videos requires replacing the original video image and the replaced face image.

Tang Hui, an artificial intelligence practitioner, said that the realization of face changing technology has low technical requirements. "As long as you're a techie who can find the right model on the Internet and understand it, then you can train that model," he said. If you just collect image data of this person and generate a video, you can train the technique in just 20 minutes. ”

When searching the mobile app store, enter "artificial intelligence face changer" to find many related applications. Among them, FacePlay is one of the most downloaded iOS apps, which has been downloaded and used by 240,000 people. The software offers many templates, including film and television characters, portraits, and comics, among others. By paying for templates, FacePlay became profitable. Members pay $17 per week and $398 for the full year. In addition, there are software that can only use its creation features by watching ads.

The following images are screenshots of the FacePlay download and premium interface.

Although e-commerce platforms have banned sensitive words such as "face changing", you only need to search for other related words, you can still find content related to them, worth up to tens of yuan.

An e-commerce platform sells AI face changing products, and the detailed pictures are taken from screenshots of web pages.

Short video platforms such as TikTok have also launched simple artificial intelligence face change templates, such as facial effects, male and female transformations, etc. A reporter tries to import a video within seconds of importing a photo, but the accuracy of this video is relatively low, and sometimes there may be some problems that do not correspond to reality.

In addition, many large technology companies are also actively deploying related businesses. Wondershare Technology (stock code SZ300624, stock price 136.60 yuan, market value 18.809 billion yuan), which has the title of AI concept stock, has developed a number of AI capabilities, such as AI face change, AI statue, AI noise reduction, AI audio restructuring, etc. According to media reports, on May 25, Wondershare Broadcasting, a subsidiary of Wondershare Technology, officially launched an AI digital human customization service, which users can use to customize digital human images, sound reproduction, and video template production. Only about 6 minutes of video and 20 effective audio materials need to be recorded to generate personalized "live" digital avatars and "real voice" footage.

At present, Xiao Zihao pointed out that artificial intelligence can produce a large number of videos in batches, replace the same character with multiple videos, or make videos of multiple characters at the same time.

AI tagging, a game of "cat and mouse".

Tang Hui believes that the rapid development of technology, especially in the field of artificial intelligence, may only see the surface today, and the technology in the next six months or a year may be completely different. However, the lack of understanding of artificial intelligence in real life has led to some fraud and misunderstandings.

Xiao Zihao said that deep synthesis technology is constantly evolving, and the generated sound and video have become more and more realistic, so it is becoming more and more difficult for ordinary people to identify with the naked eye.

According to media reports, Microsoft CTO Kevin Scott revealed in an interview on the eve of the Build developer conference on May 23, local time, that Microsoft has studied a three-year "media source system" that can add an encrypted watermark to the content generated by artificial intelligence, and then decrypt the watermark through software to obtain source information in order to facilitate the detection of false information.

Xiao Zihao introduced at the technical level that they have been exploring the method of automatic detection deep synthesis technology. Common methods include using a fake content dataset to complete the training of the model detector and identifying fake content based on inconsistencies between frames. These methods are up to 99.9% accurate in open-source datasets.

The challenge of prevention lies in the emergence of new counterfeiting methods and the increasingly complex environment in which the network is transmitted. In addition, the detection algorithm based on deep neural network has structural defects, and the anti-depth pseudo-detection technology is also facing "strong adversarial", so it needs to be continuously updated and iteratively optimized.

Xiao Zihao said that the process of deep synthesis and testing is similar to a "cat and mouse game", which requires continuous learning of offensive and defensive techniques to avoid confrontation with old technologies. In the future, in terms of anti-deep fake technology, in order to gain the initiative, we need to integrate multi-modal content forensic analysis, digital watermarking and other technical means to achieve accurate identification of counterfeit content and build a reliable content system.

The Regulations on the Administration of Deep Synthesis of Internet Information Services stipulate that enterprises engaged in deep synthesis services shall adopt technical means to add logos to the information content generated or edited by users using their services to protect users' rights and interests. For services that change the function of information content, they should be conspicuously identified in the reasonable location and area of the information content, and the synthesis of the information content should be prompted to the public to avoid public confusion or misunderstanding. At the same time, on the premise of ensuring the user experience, strengthen the management and standardization of information content, ensure the authenticity and accuracy of information content, and improve the quality and credibility of information services.

Recently, several video platforms have begun to mark the logo of AI synthetic videos, among which Bilibili has been clearly marked on some videos. And on May 9, TikTok announced platform specifications and industry initiatives on AI-generated content. It is recommended that all providers of content generated using artificial intelligence technology prominently identify the content for public judgment. At the same time, in order to facilitate the identification of other platforms, unified artificial intelligence is used to generate content data standards or metadata standards.

Station B uses AI technology to synthesize logo pictures, and the source of the picture comes from the screenshot.

Xiao Zihao suggested that if ordinary people have questions, they can consciously guide the other party to do some actions when watching the video, such as shaking their heads vigorously or opening their mouths. If the crooks are not skilled enough, they may be spotted, for example, to spot facial flaws or tooth defects, and identify AI-powered face swaps. But at the same time, he also pointed out that this method still makes it difficult to identify "high-level" scammers. In addition, you can also ask for some private information that only you and the borrower know to verify each other's identity.

Replacing a face image with someone else's image is an AI technique, but there are certain risks associated with using the technology.

According to the relevant provisions of the Copyright Law, if a performer's video material is used to change his or her face by artificial intelligence, his copyright may be infringed. In this case, the perpetrator used Mr. Guo's friend's WeChat and artificial intelligence to change his face to defraud, which was illegal use, which not only infringed the performer's copyright, but was also suspected of committing a crime.

If the AI face swapping service is used for live streaming, the live broadcaster, platform and technology provider will need to bear certain responsibilities. In particular, according to the Regulations on the Comprehensive Management of Internet Information Services, if the service providers and technical supporters of Deepin Integrated Services violate the regulations, the relevant departments will punish them; If the violation constitutes a crime, there will also be corresponding criminal liability.

Wang Rongmei's legal team from Beijing Jingshi Law Firm reminds everyone that there is still controversy over the copyright ownership of content produced by artificial intelligence. The source of the data and the algorithms used in the training process may involve intellectual property issues of others. This requires all sectors of society to jointly explore and solve the problem to provide legal guarantees for the development of artificial intelligence. At the same time, artificial intelligence will be guided to give play to its advantages on the basis of abiding by intellectual property laws and promote scientific and technological progress and social development.