Tiktok's Bold Glamour beauty filter worries AI experts about reality - The Washi...

Bold Glamour, which has been downloaded more than 16 million times since its release last month, offers one of the most convincing effects that any person with a smartphone can use to create an enhanced version or avatar of themselves in real time, say independent filter creators, researchers and creative technologists.

Unlike other photo filters, Bold Glamour doesn’t falter when there’s too much motion or if a hand passes in front of the user’s face. It alters faces in real time, and there are no stray fake eyelashes. Users have been pulling their cheeks and tugging their eyebrows because they can’t believe what they’re seeing.

TikTok spokesperson Alexa Youssefian declined to discuss the technology behind the filter, but users and researchers say that the filter reads people’s skin tones, perceived gender and hair color, then determines the amount of makeup, lip fillers and skin tightening to apply for a certain SoCal influencer look. And AI may be driving that process.

“It looks like the filter used Kylie Jenner’s face as a model for a machine learning algorithm, and then blended her face onto mine,” Laura Gouillon, a social media filter creator, told The Washington Post.

The effect is probably teaching itselfmakeup techniques, she said. “In previous beauty filters, a face mesh was overlaid onto a user’s face, so if a hand or hair covered the face, the effect glitched,” Gouillon said. “This filter is probably using machine learning technology to blend these features onto your face.”

Hany Farid, a computer science professor at the University of California at Berkeley, said the filter is probably using “generative AI,” a technology that studies zillions of pictures and text samples, often scraped from the web, to create new images or words. (Farid was a member of TikTok’s content advisory council from 2020 to 2022.)

Luke Hurd, who creates Snapchat and Instagram filters, tweeted that the Bold Glamour effect is using machine learning.

“They use something called Generative Adversarial Network,” he said. “They take an image of the user and then compare it to a data set of other images and then redraw your face, pixel by pixel, on the output of your camera feed.”

The filter’s accessibility is what sets it apart from previous means of visual manipulation, said Halsey Burgund, a creative technologist in residence at the MIT Open Documentary Lab.

How to try the new AI tech everyone is talking about

“The capacity to believably manipulate yourself, and soon others, in videos has become commoditized [through this effect], and that is the real jump,” he said. “This tech has lived in labs, companies and in more esoteric settings on the web for a while, but now everyone can do it for free and see how it works.”

Videos with Bold Glamour can be posted across the internet without a label signifying that the user has applied the hyperrealistic effect, and researchers warned that it will get increasingly more difficult to believe what’s real online.

“The Bold Glam effect is a sign that very soon, technology that creates deepfakes can also be mainstreamed,” said Memo Akten, a professor of computational art and design at University of California in San Diego. “That doesn’t mean that tech will be released, just that it’s a sign it can be released.”

Deepfakes are videos or other media that use artificial intelligence to show people doing things that never really happened.

Big Tech was moving cautiously on AI. Then came ChatGPT.

Bold Glamour videos on TikTok are disclosed with a tag directly on the video for user transparency, but that tag disappears if the videos are moved outside the app.

Company spokesperson Youssefian declined to address concerns about the warning label dropping off videos that are exported to other apps. “TikTok can only speak to users’ experience on TikTok,” she said.

TikTok parent company ByteDance has been under scrutiny for years over concerns that the Chinese-owned company’s data collection could pose a security threat.

Recently, the U.S., Canadian and other governments have mandated that federal employees delete TikTok from government devices, citing concerns about potential surveillance. At least 28 U.S. states have prohibited the app on government devices because of unspecified risks to sensitive and confidential data. A bipartisan group of senators introduced a bill Tuesday that would give the U.S. Commerce Department powers to ban TikTok or other apps based in foreign countries.

On Wednesday, White House press secretary Karine Jean-Pierre said the Biden administration has “concerns” about national security as it relates to ByteDance.

Aside from the technological concerns the filter has raised, Bold Glamour can also contribute to feelings of inadequacy and unrealistic expectations in users, especially teens and other young people, said Linda Charmaraman, a senior research scientist at the Wellesley Centers for Women.

The poreless, airbrushed beauty standard it’s pushing is far from the real-world norm without thousands of dollars’ worth of plastic surgery and makeup.

“Beauty filters used without supervision and then internalized by younger people are worrying,” Charmaraman said. “The Bold Glam filter is very realistic, and it’s less obvious that it’s a filter. And that’s why it could be detrimental for the self-worth of younger users, especially women.”

People have posted shocked reactions about how “scary good” the Bold Glamour effect is; some said the effect makes them look like they should audition for modeling, and others joked that the filter looks so real that it should be illegal. “I look like a completely different person,” one TikTok user wrote.

Apps have been enhancing faces and filtering appearances for years, but it seemed “playful,” said Akten. The Bold Glamour effect, she said, feels “ominous.”

“It’s a step toward the world we have seen in sci-fi movies where we can no longer tell what is real,” Akten said. “As a result of that, we will no longer know what to believe or who to trust.”

Heather Kelly and Tony Romm contributed to this report.