The author as Leonardo DiCaprio in ‘Romeo + Juliet’, courtesy of new face-swap app ZAO
The author as Leonardo DiCaprio in ‘Romeo + Juliet’, courtesy of new face-swap app ZAO

BusinessSeptember 5, 2019

Deepfakes, face-swaps and the future of identity: Why the ZAO app went viral

The author as Leonardo DiCaprio in ‘Romeo + Juliet’, courtesy of new face-swap app ZAO
The author as Leonardo DiCaprio in ‘Romeo + Juliet’, courtesy of new face-swap app ZAO

Earlier this week, a Twitter thread demonstrating the power of new face-swap app ZAO attracted tens of thousands of retweets. The user behind the thread, Auckland artist and game developer Allan Xia, explains what ZAO is, and what it means for your rights over your own likeness.

Last Friday, a face-swap app called ZAO went live on App stores in China and immediately took the country by storm. Days later, it would take a Kiwi-Asian Leonardo Dicaprio to spark a new global discussion around the future of Deepfake technology.

ZAO’s premise was simple: upload a photo, and in seconds users can swap faces with their favourite celebrities from iconic films, TV, music videos and more. The app was an instant smash hit, quickly climbing to the top of download charts in China where it has since firmly remained.

On Sunday evening, after seeing countless ZAO clips on Chinese social media all weekend, a bored artist down in New Zealand would finally download the app to see what the fuss was all about.

It happened in a flash: A single selfie, 8 seconds of processing, and my Asian DiCaprio was born.

Little did I realise the above video would go on to amass over 2 million views on Twitter and help reignite the global conversation about ‘Deepfakes’ first brought to mainstream attention by celebrity face-swaps back in 2017.

For the time being, I was more distracted by ZAO’s built-in meme gif generator.

And face-swapping with a CG character from one of my favourite videogames.

Not enough Asian representation in media? No problem, I could fix that.

I was in awe of what ZAO could do, and while I was extremely excited from a technologist/creative perspective, I was also wary from an ethical one. Surely, soon corporations would start advertising to us using our own faces? How many of us could resist subscribing to Disney+ if we could all become the next Avenger?

Or if we could share a stage with our favourite Kpop star?

And what about the privacy and protection of my uploaded selfie and facial data? Turns out users in China had wondered the same: just as ZAO’s viral momentum seemed unstoppable, keen-eyed Chinese netizens discovered that the app’s user agreement essentially granted ZAO complete ownership to edit and distribute all user submitted data. Boycotts and plummeting app store ratings for ZAO soon followed, as a serious debate around the dangers of privacy violation unfolded across China.

And that’s where I came in. After the app’s release in China I took to Twitter to see what the international public response was. There I was faced with a surprise: due to what I assumed to be a reverse effect of the Great Firewall of China, it seemed as if no one in the West was actually aware that ZAO existed at all.

When I started a thread to share my discoveries little did I know that I was poking the Twitter beast with the ‘China + Deepfake’ stick. What immediately followed was tens of thousands of retweets expressing excitement, shock, fear, scepticism – and also full-blown conspiracy theories about the people behind it.

So where did ZAO come from, exactly? The app was incubated by Momo, creators of a popular Chinese messaging app turned streaming platform. Facing a dwindling userbase, the development team finally hit success with ZAO, their bid to commercialise Deepfake in app form.

The problem is that now the initial buzz has faded, ZAO has faced near universal condemnation of its intrusive user agreement from Chinese netizens, and private and state news media outlets alike. Over the weekend, Alipay, China’s leading digital payment service, had to publicly reassure its users that deepfakes from ZAO would not bypass its facial recognition security checks. WeChat, the country’s largest social media platform, also swiftly blocked sharing of videos from ZAO.

By Monday, ZAO had caved to pressure and amended its user agreement so that content generated on the app will no longer be used for other purposes without the user’s prior consent. A deletion functionality has also been added, with a special notice stating if a user chooses to delete content it will also be removed from ZAO’s databases.

However, much to the ire of Chinese netizens, current ZAO terms still state that users are the ones who bear all legal responsibilities for copyright issues arising from uploaded content. Whether this is actually legally enforceable remains to be seen. ZAO seems to be counting on the immaturity of legal systems worldwide when it comes to responding to the rapid commercialisation of Deepfake technologies and the evolution of viral internet culture.

Meanwhile in the West, after the recent controversies surrounding FaceApp, many Twitter users voiced immediate concerns that ZAO is a front for the Chinese government to harvest facial data (Spoiler: it’s not very likely).

ZAO is simply a very clever use of existing technology to put users ‘inside’ their favourite film or performing with their music idols (and of course, for meme generation). Any A.I. training of facial data is limited to the preexisting clips to select from; user photos are quickly mapped on top of the original in a manner not dissimilar to Snapchat filters.

There are many techniques involved with the popular term we’ve come to know as Deepfakes – some more powerful than others at blurring the line between reality and fiction. However, at the end of the day, just like Photoshop was for photography, A.I. machine learning is a tool which is capable of creating both great art and great harm. It’s important to continue informing ourselves as technology develops, so clear-headed conversations can be had about moral/ethical boundaries and potential legal regulation.

Consider this: we’ve all comfortably shared memes of other people, but will we be as accepting when it’s millions of unfamiliar faces on our own bodies? I believe a global discussion needs to be had around ‘Fair Use’ and ‘Transformative Work’ laws to protect both creators and subjects.

When it comes to protecting our privacy, perhaps the bigger cause for worry is not so much the rights associated with what we upload online, but our likeness and identity. With a real machine-learning Deepfake, once the A.I. training is over, the base assets aren’t needed anymore and anything generated is technically an original distinct creation.

As the Streisand Effect has shown time and time again, whether it’s corporations, public figures or the average person, once something becomes a meme, it is forever part of internet culture and often essentially public domain. In the future, can we really protect what we consider dear to our personal identity?

As my ZAO videos went viral, I somehow inadvertently dethroned my hero Nicholas Cage as the current public face of Deepfake. In an ironic twist, countless top global news organisations have since used Asian DiCaprio to warn against the dangers of identity theft and privacy violation. Very few have bothered to approach me (nor, I assume, Leo) for permission first.

At the end of my ZAO story, Deepfake technologies still remain both exciting and frightening. As society stumbles into this unknown future, perhaps Asian Rose accurately captures how we all really feel:

Keep going!