Getting a tattoo is an assertion of autonomy over one’s own body. While it may not seem particularly profound when you’re in a parlour choosing humorous designs with friends, the choice to permanently alter your appearance is a conscious decision. Recently, a theory has emerged online suggesting that tattoos provide another form of bodily autonomy: protection against the replication of your image in videos created by artificial intelligence (AI) tools. It is claimed that being covered in ink offers a defence against deepfakes since current AI programmes struggle to replicate intricate artistic details. With advancements in fixed-image generation, tattoos are no longer a barrier for these technologies. A creator known as Ren stated in a viral clip, “The more tattoos you have, the less believable your imitation will be, as they cannot accurately reproduce tattoos in every frame.”
The Appeal of Tattoos as a Shield
This premise is particularly intriguing, especially among tattoo enthusiasts, who often exhibit a greater scepticism towards AI than the general public. The rise of deepfakes poses significant risks, from making it appear as though your cat has finally mastered the Macarena to being used for malicious purposes such as political propaganda, misleading advertising, and child sexual abuse material. Additionally, it would reflect an elegant historical truth: as noted by tattoo artist Jaylind in a video, cultures around the world have employed tattoos for millennia as a form of protection against physical and spiritual ailments.
AI researchers I consulted confirmed that current AI models do struggle to recreate tattoos in videos—at least, at this particular moment. Hany Farid, a professor of computer science at Dartmouth College and an expert in forensic computing and deepfake detection, noted, “There’s some truth to this, but it’s not entirely accurate.” He elaborated that while AI has advanced, it excels primarily at replicating simple visual elements. “It’s not that AI struggles with tattoos per se; it’s that it has difficulty capturing details,” Farid explained.
The Challenge of Visible Tattoos
Moreover, there are fewer individuals with visible tattoos compared to those without; only 22% of the US population has more than one tattoo. This means that the training dataset for models featuring tattooed individuals is considerably smaller than for those without tattoos, according to Saetbyeol Leeyouk, a critical technologist and graduate student at MIT Media Lab who focuses on the intersection of ethics and AI.
“If you claim that ‘tattoos make it difficult for generative AI to recreate your image,’ you must conclude that statement with ‘today.’ Tomorrow, next week, next month, things will improve. We know this,” Farid explains. Leeyouk concurs, stating, “Over time, the issue of tattoos will be resolved.”
The Evolution of AI Capabilities
As evidence of this acceleration, Farid recalls a 2023 video in which Will Smith appears to eat spaghetti, a nightmarish montage featuring figures vaguely resembling him shoving pasta-like balls into what could be mistaken for a human mouth if one were unfamiliar with the species. Three years later, a Reddit user created an updated version that, unless one scrutinises too closely for a human soul behind the eyes, could very well pass for a spaghetti commercial starring Will Smith.
These AI-generated videos are improving as programmes refine temporal coherence, which refers to the visual consistency between frames. “Temporal coherence is achieved because we no longer generate frame after frame after frame,” Farid explains. However, there are still limits concerning the length of these clips; typically around 10 to 15 seconds. For now, your tattoos may appear distorted in any longer AI-generated video. “If you have many tattoos and a video is made of you, they would likely shift slightly from one clip to another,” he asserts.
Understanding the Implications
But how much does that slight movement truly matter? Leeyouk points out that, particularly in the case of non-consensual sexual deepfakes, the objective is not just the video itself but to humiliate the victim: “You probably won’t care much if they’ve nailed the shading on your neo-traditional snake if your image is being stolen for malicious purposes.”
So, is there anything that can protect you from becoming a victim of a deepfake? “Build a bunker and disconnect from the world. Five years ago, it was the Scarlett Johanssons of the world who were vulnerable to deepfakes, as thousands of images and hours of voice recordings were needed. Today, it takes just one image of you or 10 seconds of your voice, and that’s it,” says Farid, who resides on 100 acres of woodland in rural Vermont. Even if you don’t post much online, someone with Meta’s smart glasses could capture much more than that simply while you walk down the street.
The Risks of False Security
Leeyouk is concerned that videos suggesting tattoos provide any form of protection may lead people to develop a false sense of security. “I know people are actively seeking ways to protect themselves against deepfakes. But we should be pressing politicians or technologists to take responsibility for this, rather than looking for ways individuals can protect themselves,” she argues.
When I asked if there was any reason for optimism in this regard, Leeyouk replied that as AI programmes advance, so too will the tools for detecting AI-generated content. “If detection techniques improve, it means we will be able to easily distinguish which are AI-generated and which depict real victims,” she explains.
So no, there’s no need to rush out and get a multitude of tattoos to shield yourself from deepfakes: the “protection” they offer is likely to fade soon. “There are many good reasons to get tattoos. But they won’t save you from AI,” Farid concludes.