ByteDance, the owner behind TikTok, has been secretly building a deepfake feature that would allow users to insert their faces into videos they’re not actually in, according to a TechCrunch report. While the feature is not live yet, it would appear that TikTok and Douyin—TikTok’s sister app in China—both feature code for taking multiangle biometric scans of a user’s face. The scan can then be inserted in select videos before being shared.
The news sounds concerning, especially when you consider how manipulated media can potentially be misused to spread misinformation. See: That faked viral video of Nancy Pelosi appearing to slur her words. TikTok has also been the subject of scrutiny as of late, with accusations it collected children’s data, military bans, and national security concerns.
The code itself was discovered by Watchful.ai, an Israeli startup. Their findings indicate that ByteDance is hyper-aware that a potential deepfake feature might not be received too well. As a result, it appears ByteDance has gone through the trouble to build in some safeguards. For instance, users must scan their face from multiple angles to create the deepfakes—a measure which doubles as identity verification and a preventive measure from someone using a single photo to create manipulated media without consent. Users can also only insert themselves into a limited number of videos that ByteDance claims to have rights to. Lastly, when it comes to sharing, the generated videos will feature watermarks to indicate the content isn’t real.
Watchful.ai also found unpublished updates to TikTok and Douyin’s terms of service with regard to the deepfake feature. It reiterates that “real identity verification” is required and that uploaded photos can’t be used. Also encouraging is the fact the feature can’t be used by minors.
Neither TikTok nor Douyin seems keen to own up to the feature existing, however. Spokespersons from both apps denied aspects of Watchful.ai’s findings to TechCrunch. Specifically, TikTok denied it was an upcoming feature but also seemingly inadvertently acknowledged the code existed by stating “The inactive code fragments are being removed to eliminate any confusion.” That disconnect between the two apps tracks with recent reports that ByteDance is doing its best to isolate TikTok from the rest of its Chinese operations over espionage concerns. In an email to Gizmodo, a TikTok spokesperson said, “TikTok is not experimenting with this feature, has never offered this feature, and has no intention to offer such a feature in the future.”
It’s hard to say whether the deepfake feature will ever see the light of day in the U.S. It’s possible that it could be a China-only release, or that ByteDance might find launching a deepfake feature isn’t worth the potential blowback. Even so, ByteDance isn’t the only social media app to dabble with media manipulation via filters or face swapping. Snapchat, for one, has had a face swap feature for years now. Chinese app Zao took it a step further, allowing users to transpose their faces onto famous actors in short clips. The problem is the more sophisticated these deepfake tools get, the murkier it gets with regard to IP, misinformation, and potential abuse. Whether any other apps attempt to incorporate more extensive deepfake tools into their platforms remains to be seen.
Update, 01/03/2020, 1:46pm: Added comment from TikTok spokesperson.