Two First-of-Its-Kind 'AI Face-Swapping' Software Infringement Cases Pronounced in Beijing Internet Court

TapTechNews June 20. According to the Beijing Internet Court, on June 20, the Beijing Internet Court first instance held a court session and pronounced judgments in two first-of-its-kind "AI face-swapping" software infringement cases in Beijing, and it was determined that the online service provider who used others' videos for "face-swapping" to make templates and then provided "face-swapping" services violated the personal information rights of others.

Both plaintiffs Liao and Wu are traditional Chinese style short video models and have a large number of fans across the whole network. The defendant is the operator of a "face-swapping" App.

Plaintiffs claimed that without their authorized consent, the defendant used the plaintiffs' on-camera videos to make face-swapping templates and uploaded them to the involved face-swapping App, providing them for users to pay and use to make a profit.

Plaintiffs believed that the defendant's actions violated the plaintiffs' right of portrait. At the same time, the defendant uploaded and used videos with the plaintiffs' portrait information without the plaintiffs' consent, which was the defendant's illegal acquisition and alteration of the plaintiffs' face information by extracting the plaintiffs' faces through AI technology and replacing them with the faces of a third party, and then using the technologically processed videos as paid templates for users of the involved App to use and make a profit from, which violated the plaintiffs' personal information rights. Plaintiffs therefore demanded that the defendant apologize and compensate for moral and economic damages.

The defendant argued that the videos released on the defendant's platform all had legal sources, and the facial features were not those of the plaintiffs, and there was no violation of the Plaintiffs' right of portrait. In addition, the "face-swapping" technology in the App was actually provided by a third party, and the defendant did not process the Plaintiffs' face information and did not violate the Plaintiffs' personal information rights.

After the court hearing found that the defendant did not submit evidence to prove the source of its template videos, combined with the fact that the characters' makeup, hairstyle, clothing, movements, lighting and camera switching in the template videos presented consistent characteristics with the plaintiffs' on-camera videos, it could be determined that the defendant used the Plaintiffs' on-camera videos and replaced them with others' faces through deep synthesis technology and then uploaded them to the involved App as templates for users to use. However, this action did not violate the Plaintiffs' right of portrait.

The court held that the defendant's actions constituted an infringement of the Plaintiffs' personal information rights:

First, there was personal information of the Plaintiffs including the Plaintiffs' face information in the Plaintiffs' on-camera videos. The Plaintiffs' on-camera videos dynamically presented the individualized characteristics such as the Plaintiffs' facial features, and based on digital technology, these individual characteristics could be presented in the form of data, which conformed to the definition of "information related to an identified or identifiable natural person" stipulated in the Personal Information Protection Law of the People's Republic of China.

Second, the defendant implemented actions to process the Plaintiffs' personal information. Firstly, the defendant should be the subject responsible for processing personal information. Even if the defendant actually used the technical services of an outside company, the outside company was only the entrusted technical service provider, and the defendant was the principal of personal information processing, determining the way and scope of information processing, and should be responsible for the personal information processing behavior. Secondly, the face-swapping action in the involved case belonged to personal information processing behavior. The defendant first needed to collect the Plaintiffs' on-camera videos containing the Plaintiffs' face information, and replace the Plaintiffs' face in the video with the face in the provided photos. This process used the face recognition technology of detecting key points of the face, and then fused the face features corresponding to the provided face image onto the specific characters in the template image, generating a picture with both the face features of the specified image and the template image. The synthesis process was not only a simple replacement but also needed to fuse some facial features and expressions of the original video with the features of the new static image through an algorithm to make the replaced template video appear natural and smooth. The above process involved the collection, use, and analysis of the Plaintiffs' personal information, so the process of forming a face-swapping template video through "face-swapping" belonged to the processing of the Plaintiffs' personal information.

Third, the defendant's actions violated the Plaintiffs' personal information rights. The processing behavior of automated personal information often has characteristics such as concealment, thus the law prevents the risks of leakage and abuse by granting individuals the right to know and decide on the processing of their personal information. Although the Plaintiffs' on-camera videos belonged to already public videos, there was a note on the involved account stating "not authorized to any paid software", and it should not be presumed that the Plaintiffs agreed to the processing of their face information by others. In addition, the defendant obtained videos containing the Plaintiffs' face information, analyzed and modified them using this emerging technology of deep synthesis, and then commercialized them, which might have a significant impact on the Plaintiffs' personal rights and should be obtained with the consent of the Plaintiffs in accordance with the law. The defendant had no evidence to prove that it had the Plaintiffs' consent, so it constituted an infringement of the Plaintiffs' personal information rights.

The court ruled that the defendant apologize to the Plaintiffs and compensate for moral damages and economic damages. Currently, the case is still in the appeal period, and the first-instance judgment is not in effect.

Judge Sun Mingxi said that the two cases pronounced this time are new types of cases involving "AI face-swapping". The special thing about these cases is that the defendant uploaded the Plaintiffs' videos after face-swapping them to the application software. Whether it constituted an infringement of the Plaintiffs' rights and what kind of rights it was, these were new problems brought about by the development of new technologies. In this case, the defendant essentially exploited the Plaintiffs' short video, and this behavior included the use of a part of the Plaintiffs' face information for the fusi on of the newly uploaded photos and the use of factors such as the makeup, hairstyle, clothing, and overall styling as well as the lighting and camera switching in the video to form a template video. In other words, the main factor for the defendant's profit was the labor input in the Plaintiffs' involved video. If the defendant used the above elements without the consent of the relevant rights holders, "free-riding" on the labor input of others, the relevant rights holders could safeguard their legitimate rights and interests based on other claim right bases such as labor creation input and competitive interests. The Plaintiffs did not claim to be the relevant rights holders for these rights, so the court only dealt with the part of property damage compensation for the infringement of the Plaintiffs' personal information rights by the defendant in this case.