It was 6:28 am when I woke up to a text from a friend in Shanghai, China.
“Hey, Amanda—is this you?” he wrote via WeChat.
I hadn’t even had my morning coffee yet. I pulled my phone closer to get a better look.
“Yes, it’s me,” I typed back. “But … how?”
While scrolling through Taobao, a Chinese marketplace owned by Alibaba, my friend came across an ad for a camping stove. It was like looking in a mirror—I saw my Puerto Rican mother’s long eyelashes and distinct jawline, my father’s prominent Austrian nose, and my abuela’s long hands.
“Is it Photoshop?” “Was I hacked?” “Or perhaps one of my photo apps is to blame?”
All plausible questions. Having lived in China for a few years, I was used to bystanders snapping a photo or two, as this is quite normal upon seeing foreigners, especially redheads. But I had never given my consent, let alone posed for this photo. Come to think of it—I didn’t even own a white winter jacket.
So I started investigating.
First, I used a slew of tools—TinEye, Google, Bing, Yandex, and others—to reverse-search the photo across international sites, eventually saving each copy in a single desktop folder. In total, I found the photo reposted 74 times on marketplaces from Germany to Japan. The products and details changed in the photos—some advertised a camping stove while others featured a portable gas cooker. And while it was clear the images derived from the same source image, one photo from a marketplace in Belize stopped me in my tracks. This version was wildly different. Sure, she looked like she could be related to me, but I knew she wasn’t me due to her slightly rounder face and a small gap in her teeth, which I don’t have. Unlike me or any of the versions that appeared on Chinese sites, she also had eyes that were sunken in, and her cheeks appeared as if they had been layered a few times.
Eventually, after cropping the photo and plugging it back into reverse search tools and even a few catfishing sites, I found the source image: an Amazon ad for an outdoor camping tent. The original Amazon marketplace model slightly resembled me, but more like she was a cousin. (“We looked into this and have confirmed the photo on our site was taken in 2018, and the model in question is an Amazon employee,” said Betsy Harden, a spokesperson for Amazon.) The more the image was reposted, from site to site around the globe, however, the more it shifted and transformed to include elements of me and my likeness.
The Amazon model resembled the Belizean version, but the mouth was not nearly as wide, and the original photo overall had a bit more softness to it. In the version that ended up on Taobao and JD—the one that looked like me to the tee—the jawline was sharper, the lips plumper, the face smaller, the chin more elongated, cheeks rosier, and brows more defined like mine.
Rijul Gupta, a synthetic media engineer whose company DeepMedia AI recently partnered with the US Department of Defense to flag deepfakes, said the Chinese marketplace photos could have been created using an actual photo of me. If so, the images were likely manipulated not with Photoshop but with deepfake synthesization tools, which is a complex process but significantly faster than the former. These sophisticated tools allow you to take any face you find online and manipulate it according to your needs.
“There is clear evidence that the images in question have been synthetically manipulated,” Gupta told me. “This is evidenced by clear artifacts in the photo around Amanda’s face, improper lighting, and highlights on the clothing.” In the Belizean version, “these checkerboard artifacts around the chin are a dead giveaway” and are also distinct from any sort of image compression you’d normally see.
It’s unclear if marketplace sellers manipulated and synthesized my face intentionally—which can be seen on Douyin (Chinese TikTok), Weibo, and other social media sites—or whether they pulled from “random” photos online to create the synthesis. “In either case, this shows the clear threat this technology poses to those who may suffer from identity fraud,” Gupta said.
To learn more, I reached out to an AI engineering team in Hong Kong. Wendy Zhang, a senior engineer for Cutout.pro, ran tests on the ads from Chinese marketplaces. For deepfakes, the system would give it a value of 1, whereas genuine photos would be assigned a 0.
No surprise—there were 1s across the board, with the exception of one photo. The team tested multiple versions of the photos, including close-up images of the altered face and hands, as well as genuine photos of me that I provided for comparison.
“The algorithm keeps learning by itself,” Zhang said. “So the deepfake images evolve over time and change or become better in various ways.”
Hijacking the System
There are usually two reasons people produce images like this—one being that they’re trying to sell products and don’t want to pay for an original image, the other being that they want another face in the picture.
“It would make sense for some companies to ‘hijack’ related Amazon images, insert their products into it, and use the modified image for their own marketing,” Neil Sahota, an artificial intelligence adviser at the United Nations, said. “It’s a major shortcut and would probably also explain why they don’t replace the original models with local people.”
“For the cooker ad, it is possible they incorporated some of Amanda’s likeness,” he added. “They would need to alter the photo just slightly to fool Amazon of any image copyright duplication.”
It’s not like there’s a shortage of fresh faces or talent, so why do people do this?
“They think your face sells better than the original model’s,” Gupta said. “What will happen is they’ll take an image, and it goes from person to person, modifying the face the entire time.”
Depending on the quality of the deepfake network, you may be able to achieve what he calls a “perfect face swap,” which matches a person’s identity. Other times, it will be 50/50. “It’s going to continue down that cycle with different identities and different faces the entire time,” he said.
Face swapping and deepfake tech aren’t new. Apps like Face Juggler and Face Swap Live made the rounds in 2012 and 2015, but since then, these tools have become more sophisticated, allowing someone to easily change a model’s body position and clothing too.
Among the most famous examples is the video of Ukrainian president Volodymyr Zelensky, which hackers created to make it seem like Zelenskyy was surrendering to Russia. When Olympic gold medalist Liu Xiang’s likeness was used for commercial purposes, the court ruled in Liu’s favor, awarding him 6,000 RMB ($872). Oh, and there’s the uber-famous and controversial Bruce Willis deepfake commercial made for a well-known phone carrier in Russia. In the commercial, the fake Willis and another actor are tied to a bomb that’s set to go off if they don’t diffuse it in time. The stunt opened up conversations about a person’s right to their face, name, and likeness.
As for my case, the marketplace sellers did not reveal how the ads were constructed, but Taobao, AliExpress, and JD sellers have since removed the images in question. The other images still live on a few Chinese blogs and on the Belizean site.
Copy, Paste, Repeat
Copy-and-paste culture has plagued the economy for years. Knockoffs and counterfeit goods cost the global economy over $500 billion every year, according to the US Chamber of Commerce.
“People say, ‘Good artists copy, great artists steal.’ The copycat behavior is essentially driven by learning,” Howard Yu, a professor at IMD business school in Lausanne, Switzerland, said. “They try to eke out a living by copying what others do; however, this often does not lead to success.”
During my investigation, models, and influencers messaged me to share similar stories of identity and likeness infringement, particularly by companies in China that reuse photos for future campaigns without giving models additional payment or credit.
“I had a buddy send me a picture of myself on a huge billboard on the side of the road for a pretty big brand, and I did not do a shoot for that brand,” South African model Jay d’Engle, 32, who lives in Shanghai, said. “Then it hit me, ‘Ah—that one casting.’”
In most cases, casting calls are held so agencies can select models for a shoot and test out a few looks, but some turn these scouting events into full-on photo shoots that are later illegally repurposed. “Seeing these kinds of things happening again and again—I feel hopeless, and I don’t know what to do anymore,” Lynn, an influencer with a cumulative 1.5 million followers online, told me via WeChat. “I feel my country is not really helping me with that.”
Lynn’s fans discovered that her photos had been used multiple times for marketing purposes—but her face had been swapped out for another. The 27-year-old from Wenzhou, Zhejiang, frequently posts on Weibo and 小红书 (RED) and says that this type of infringement has happened at least 15 times since she began her career as a full-time blogger seven years ago.
Even if you get a lawyer involved, as some have done in the past, these cases, in abundance, can take time to resolve and often involve high legal fees because “the technology threshold for committing these infringements is low, and the legal consequences are generally not severe,” said Horace Lam, who co-leads a team of IP and technology lawyers in DLA Piper’s Asia offices. “Some infringers willfully choose to infringe in exchange for the economic benefits.”
Lynn, who remains active on social media, thinks lack of education is a big issue. “Some people don’t even know that it’s not legal to use other people’s photos,” she said. “Most people say, ‘OK, I’m just gonna use this photo—it’s not a big deal.’ And if every case were taken seriously, it would take a lot of time.”
A Legal Affair
Often, when people think of deepfakes and synthetic media, they immediately jump to the adult film industry or politics, but these synthesizations go far beyond such industries. “It’s not just in the realm of world leaders and national security anymore,” Gupta said. “It’s getting into the realm of personal identity and personal security.”
As of 2019, users in China are also required by law to be upfront about their use of deepfake-generated, AI, or VR-related media—otherwise, they could be charged with a criminal offense. Remember Chinese Elon Musk, who gained the attention of the Tesla CEO himself? Chinese Musk, known as “马一龙,” or Yilong Ma, was indefinitely suspended on Douyin, the original and separate version of TikTok, last May for failing to disclose to his fans that his videos were created with deepfake technology.
In 2020, China passed the Civil Code of the People’s Republic of China, which protects a person’s personality rights and portrait rights. Personality rights, defined by Article 990, include a person’s right to their name, portrait, reputation, honor, and privacy, among other rights. Portrait rights protect a person’s likeness, personal image, and appearance.
“That includes people using technology to fake your likeness, you know, any kind of drawing of your likeness,” Jeremy Daum, senior fellow of the Yale Law School Paul Tsai China Center, said. “You can have a civil action to sue for damages. Usually, what happens, though, is—law or no law—it’s such small potatoes in terms of damages.”
In my case, with my image or likeness circulating on Taobao, JD, and other marketplaces, I would be protected under these Chinese laws. Additionally, the Cyberspace Administration of China recently announced that all platforms in China that create or provide these types of services now need to get a person’s consent to use their voice or image in a deepfake.
So, China, as a whole, does have stringent laws in place. In the US by contrast, California, Texas, New Jersey, New York, Hawaii, and Illinois have certain deepfake restrictions in place, too, but they’re not catch-alls, and these laws still leave room for bad-faith actors to scoot around regulations. It’s time to establish stronger protections that cover international waters. When we talk about media making the rounds in the US or in China, we’re actually talking about media that’s generally accessible around the globe thanks to the internet.
Victims of deepfake technologies have no choice but to resort to specific domestic laws, Lam said. He and his team expect these issues to increase in coming years, far beyond China and the US. ”We are seeing more gray areas where laws need to catch up with technology,”he said.
Ultimately, I’ve learned that anyone can use image synthesization and digital tools to copy, paste, and repeat. Bruce Willis and I are apparently no exception.
“I think this is the next version of identity theft,” Gupta said. “And it’s just getting started.”