Popular deepfake apps are making it easier than ever to make AI-powered manipulated videos — spawning new memes, and an increased potential for abuse
Rob Lever/Getty Images
- Making deepfake videos — which use AI to convincingly swap one person's face with another's — is easier than ever before.
- While deepfakes were once the domain of sophisticated coders, new apps are democratizing the technology and making it accessible to wide swaths of people — and turning a profit while doing so.
- The trend has spawned new memes on apps like TikTok and YouTube, as well as new concerns about their potential for abuse or misinformation, according to experts.
- Deepfake app makers argue that making the technology more widely available could help educate people about how to spot the telltale signs of deepfakes.
- Visit Business Insider's homepage for more stories.
Dozens of heads bob in unison over a techno beat, their facial expressions synchronizing to match the face of a TikTok star lip synching the lyrics. As she mouths the words, videos of Barack Obama, Vladimir Putin, Benedict Cumberbatch, an Easter Island head, and the clown from "It" follow along in perfect lock-step.
Unsurprisingly, the video isn't real — rather, it's an example of an AI-generated deepfake, and part of a meme that's sweeping platforms like TikTok and Instagram.
While deepfakes have typically been seen as sophisticated technology outside the reach of amateurs, new consumer-facing apps are making it increasingly easy to generate convincing lookalike videos that appear to show a person saying something that they never actually said.
So far, the arrival of these apps has spawned memes and social media trends that are mostly harmless — but experts warn that the increasing availability of the technology could spawn a misinformation crisis.
New technology powers a deepfake explosion
Many of the latest deepfake memes were made using RefaceAI, a new app that has built its own machine learning frameworks that enable users to take a selfie and merge it with a target video or gif to create a deepfake. Unlike face masking tools found on apps like Snapchat, Reface generates an entirely new video using the twin inputs, CEO and cofounder Roman Mogylnyi told Business Insider.
"Before, making a deepfake video could take a week or more, and even then you wouldn't be satisfied with the results," Mogylnyi said. "It was a challenge for us ... we wanted people to have Hollywood level post production on their phone."
Reface was founded by a group of Ukrainian software engineers but is incorporated in the US. After piloting an early version of their technology as a web service last year, Reface debuted on the App Store and Google Play stores in January 2020.
Since then, it's popularity has skyrocketed. Memes made with Reface have been retweeted by Elon Musk and Britney Spears, and the app soared to the top of app store charts in more than 100 countries. It has been downloaded more than 42 million times, and has been used to make hundreds of millions of deepfake gifs and videos, according to Chief Business Officer Dima Shvets.
—chrissy teigen (@chrissyteigen) August 26, 2020
Reface currently makes money through advertisements and premium subscriptions on its app, but its creators envision building an entire Reface platform similar to Snap's Bitmoji or Apple's Memoji that users could use to personalize their interactions and use across social media.
The advancements are raising new concerns about abuse — and measures to stave it off
Experts have warned for years that deepfakes could pose a unique misinformation threat if they're weaponized to mislead people about the actions of an individual or public figure. As the technology becomes more widely available, those concerns are amplifying.
One of the earliest uses of deepfakes was to harass women by creating synthetic porn videos that grafted their faces onto sexually explicit content without their consent. It's also been used for political misinformation — a deepfake video that appeared to show Belgium's prime minister making false claims about COVID-19 went viral earlier this year before it was debunked.
Deepfake technology could be evolving too rapidly for public understanding to keep up, according to Nina Schick, author of a book titled Deepfakes: The Coming Infopocalypse.
"Unfortunately, I think synthetic media is so fun, the technology is so nascent, it's such an area for growth and so much private investment is backing it that it will just explode and the ethical and moral considerations will be three steps behind," Schick told Business Insider.
Reface is already taking steps to counter potential misinformation, Mogylnyi said. It has pornography detection meant to automatically remove nudity. The app currently only lets people make deepfakes using gifs or 10-second videos, and Reface is building out digital watermarking for all videos before it expands the tools available to users.
"Currently, in my opinion, we're only giving users access to 10% of the technology's capabilities," Mogylnyi said.
He added that Reface has shared its API with companies like Facebook to help build tools to detect deepfakes, and that Reface is working on its own detection tool that would let people upload media to check whether it's synthetic.
That approach was echoed by Sergey Tokarev, a lead investor of Reface who feels the technology's promises outweigh its potential risks.
"No technology can be safe from misuse in a negative way. But does it mean that technology should not be developed?" Tokarev said. "My point is that all technologies should serve the world development and be used in good faith."
Startups are focusing on teaching people about deepfakes and how to spot them
Other startups dabbling in deepfake tech are choosing to deploy it through narrow avenues in order to avoid abuse. Kapwing, a San Francisco-based startup that makes video editing tools geared towards influencers, published a tutorial that links to the code to create a deepfake synced to one specific song that became popular on TikTok, without giving people tools to make a wider range of deepfakes. The video has been viewed more than 500,000 times.
"We've been very careful about our approach to this," Robert Martin, a strategist at Kapwing, told Business Insider. "I'm reluctant for us to do much more with this technology than what we've done here."
As deepfakes become more widespread, there could be a rise in "deepfake literacy," or people's general awareness of deepfakes and ability to spot the telltale quirks inherent in the videos. But future advances in technology could make deepfakes even more seamless — and, according to Schick, deepfakes becoming commonplace could reduce people's trust in real media.
"Everything can be fake, and everything can be denied" Schick said. "The democratization of this type of manipulation of media is just one step into the future, but I don't think we're ready for it."
from Business Insider https://ift.tt/36cLnmG
No comments