Chinese “DeepFake” App Goes Viral, Renewing Concerns About Potential Misuse of Face-Swapping Tech
“DeepFake” face-swapping technology created a buzz a couple of years back when a series of fake celebrity porn videos generated by the method spread rapidly across the Internet. Deepfake’s unseemly side triggered widespread fears and concerns regarding misuse of the new tech, although its propagation was limited by the relatively advanced programming expertise required to perform convincing face-swapping.
Well, that tech threshold has now been significantly lowered. A recently released Chinese deepfake mobile application, “ZAO,” enables just about anyone to easily swap faces with popular characters such as Sheldon Cooper in The Big Bang Theory, Marilyn Monroe’s Lorelei Lee in Gentleman Prefer Blondes, Jack Dawson in The Titanic, and so on.
ZAO for iOS and Android was launched last week and became a runaway hit, with a fresh crop of homegrown face-swapping videos flooding Chinese social media sites like Weibo and WeChat. In an official statement, ZAO’s creators said they spent some two million Chinese yuan (approximately US$280,000) on servers in just one night to process the huge volume of requests.
The company that developed ZAO is owned by Momo, whose dating and messaging mobile app is known as “China’s Tinder.” Momo went public on the Nasdaq four years ago and is valued at US$7.63 billion.
To face-swap on the ZAO platform a user first uploads a portrait photo, and is then prompted to blink, open their mouth, bow their head and so on to enrich their facial information. The user can swap their face with one of the characters in ZAO’s celebrity library to generate a shareable video or GIF of up to 10 seconds. Friends can remotely collaborate on ZAO by swapping their faces into the same video.
The tech at the heart of all this face-swapping is generative adversarial networks (GAN), a powerful AI framework comprising two neural nets which challenge each other to produce increasingly realistic fakes. Even the most popular face-swapping desktop applications however, such as DeepFaceLab and FaceSwap, still require users to leverage hardware, datasets, and programming skills to extract features from videos, train models, and finalize their creations.
ZAO’s rapid rise is in large part due its ease-of-use: all you need to run the app is a smartphone and a selfie. On August 31, ZAO was China’s 2nd most downloaded iOS entertainment app and ranked 7th on Weibo’s “Hot Search” real-time list of hottest discussions based on users’ search results.
ZAO’s popularity however has brought with it concerns about the app’s user agreement, which grants ZAO, its affiliates, and its users the “free, irrevocable, permanent, transferable and licensable rights worldwide” to morph or edit their content (such as replacing a face or voice in a short video with another person’s face or voice) and disseminate content.
Facial recognition technology has been widely adopted for mobile payment systems in China, and many netizens are concerned their facial information could be misused for identity theft. A typical post on Weibo last week warned: “If your facial information is hacked, hackers might be able to access your AliPay (Alibaba’s digital wallet) in a minute.”
The ZAO user agreement also attempts to absolve the company of any liability regarding portrait rights, stating that the user bears the legal responsibility for any content they post on the platform. Jun Wang, chief partner at Beijing Ta Law Firm, told Chinese media “What ZAO means is that the legality of the user content and the authorization of the copyright should be resolved by the user. Otherwise, the user has to be responsible for all legal liabilities.” Wang noted however that despite the language in the user agreement, it cannot fully exempt the ZAO platform from legal responsibility.
This April, China’s Standing Committee of the National People’s Congress began drafting a Civil Code Personality Rights document which forbids infringing on the portrait rights of others by means of digital technology forgery.
Another related public concern is deepfake’s potentially serious threat to credibility, particularly involving fake video evidence. To assuage those fears ZAO only enables modifications to its own library of video clips.
Synced notes that last weekend ZAO removed the “free, irrevocable, permanent, transferable and licensable rights worldwide” clause from its user agreement. Yesterday, ZAO released an official statement pledging that the company will not collect or store the biometric facial information of its users.
Journalist: Tony Peng | Editor: Michael Sarazen
We know you don’t want to miss any stories. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.
Need a comprehensive review of the past, present and future of modern AI research development? Trends of AI Technology Development Report is out!
2018 Fortune Global 500 Public Company AI Adaptivity Report is out!
Purchase a Kindle-formatted report on Amazon.
Apply for Insight Partner Program to get a complimentary full PDF report