Hi, everyone. So now Elon wants to buy Twitter, allegedly to help him build X , “the everything” app. Sweet of him to name it after his kid .
For a couple days in late September, no one seemed clear on who owned Bruce Willis. The British newspaper The Telegraph claimed that the actor, who has retired because he suffers from aphasia, had digitally reincarnated his career by selling performance rights to a company called Deepcake , which used artificial intelligence technology to map Willis’ face onto another actor. Not long after, representatives of Willis said that the star of Die Hard had done no such thing and had no relationship with Deepcake, even though the company’s website had a complimentary quote from the star.
The episode raises a lot of questions, not least the meaning of identity at a time when one’s image can be so easily faked. So I went to the source and spoke to Deepcake’s founders. The two-year-old startup from the former Soviet state of Georgia is the project of Ukrainian-born CEO Maria Chmir, a marketing executive, and head of machine learning Alex Notchenko, who has a doctorate in AI.
Chmir told me that the company never claimed to own Willis’ future rights, but had a previous and mutually satisfying arrangement where Deepcake digitized his appearance in a 2021 ad for Megafon, a Russian cell network. The Willis ad is part of Deepcake’s game plan to serve customers who want to digitally clone humans. “We are one of the first on the market to be commercially successful in the field of legal deepfakes,” says Chmir.
“But we don’t like this word. These are sort of replicas, or digital twins. ” (I wondered why, if she wasn’t fond of the word, she named her company on a variation of it, but whatever.
) How good is that technology? Let’s go to the tape . In the Megafon commercial, a person who is unmistakably Willis, even if you know it really is not, is among two hostages tied to a ship mast, next to a digital clock ticking down seconds before a bomb goes off. While the figure has Willis’ face, it doesn’t quite convey his trademark insouciance.
And for some reason, this Willis has a different voice—a gruff bark that speaks Russian. Still, it looks like Willis—digitized and generated, Chmir says, by algorithms trained on 34,000 images from his earlier films. Chmir says that Willis was deepfaked because he wasn’t available to travel, but the process makes economic sense as well.
While leasing an actor’s rights might be about 30 percent less than the usual appearance fee, she says, still bigger savings come from the lower costs of filming a cheap actor-double instead of a superstar, who requires first-class travel, a big trailer, and ridiculous demands in contract riders. But Deepcake isn’t just faking superstars. They recently did a job for an agricultural firm that wanted to make educational videos starring its in-house expert, a busy person not comfortable in front of a camera.
With the subject’s permission, Deepcake converted video of an understudy in an exact duplicate. “We also cloned the voice for full similarity, of course,” Chmir says. That seems disturbing.
Don’t the people watching this stuff deserve to know that they’re watching a fake, or “twin,” instead of the real person? “Absolutely! There should be responsible disclaimers,” says Notchenko. But there was no such disclaimer on the Willis ad. Notchenko explains that no disclaimer was necessary because the fake version of Willis, who is 67, looks like he did in his younger years.
And he has a smartphone, which the younger Willis did not have access to. On the other hand, Chmir says that 80 percent of people watching the video thought it was really Willis. Deepcake is far from alone in mapping famous faces onto stand-ins.
We’ve already seen major characters in blockbuster films who never had to respond to a call sheet. Death apparently had no dominion to stop Carrie Fisher from reprising Princess Leia in Rogue One . And Peter Jackson’s Weta Digital studio reanimated the late Paul Walker to ride again in Furious 7 .
The techniques for mapping an actor’s face onto a stand-in are only getting better. In June 2020, three scientists at Disney Research Studios published a paper describing “an algorithm for fully automatic neural face swapping in images and videos. ” They claim that their system clearly vaults over the dread uncanny valley , where fakes aren’t quite convincing.
Competing with Disney doesn’t seem to faze Deepcake’s Chmir, who says the company’s technology was used in a film screened at the Cannes festival recently and is currently working with other celebrities, including Jean-Claude Van Damme. She wouldn’t say whether they are using a younger version of the 61-year-old actor and martial arts fighter. For all the focus on actors when it comes to deepfakes, Hollywood is actually the least controversial arena for this technology.
Movies have already crossed the digital rubicon where anything in a given shot could either be computer-generated or real. The real trouble from AI-generated faces will come when they bleed out into our everyday existence. In the early weeks of the Russian invasion of Ukraine, a deepfake of Ukrainian president Volodymyr Zelensky announcing his surrender appeared.
It was relatively crude, and thankfully it was quickly debunked. Just this week, the moderator of a large LinkedIn group reported a flood of fake accounts professing to be executives of real companies. Their names were plundered from obscure IMDB listings, but their profile pictures were computer-generated.
(Creepily, all were women. ) And earlier this month, WIRED’s own Lauren Goode wrote about her experience of being matched with fake men on the Hinge dating app. Her unreal suitors appeared to use stock, static photos.
But imagine what could happen if sophisticated deepfake video tech hits the dating world. We have a reality problem, and movie actors are the least of it. I’ll leave it to the lawyers to determine who owns the rights to Bruce Willis’ punim.
The tougher question is, how do I know whoever is presenting themselves to me is really them? And how are we going to protect our own images from being misused? Sadly though not surprisingly, a prime use of deepfakes to date has been mapping celebrity faces to naked people. (Trust me, the first search results for Carrie Fisher deepfakes probably do not involve Rogue One . ) And Chmir says this is only the beginning.
Just wait until improved deepfake technology meets better AI natural language generators. “When we combine all of these technologies, we can talk about digital immortality,” she says. “Because we can recreate any person, like a digital twin with that person’s mindset, behavior, voice, and appearance.
” And where will you meet those computer-generated twins? The metaverse! If our avatars move past being cartoons and become photorealistic, Deepcake will be ready . “We have a lot of requests from companies who are developing their own metaverse, and celebrities and agents who want to digitize their stars for the metaverse,” says Chmir. The metaverse will be so fake that fake will be real.
In a world where everything is made of pixels, the odds are low that your photorealistic avatar will actually resemble that imperfect creature you see in the mirror. It might look like an idealized version of yourself, like a Photoshopped model in a fashion ad. Or perhaps it will be someone cobbled together from photographs and videos captured years earlier, when you were in the bloom of youth.
Or maybe your avatar will be a perfect twin of an entirely different person, borrowed with or without permission. Maybe it will even be Bruce Willis. Yippee-ki-yay , meta-fucker! In the Winter 1991 issue of Premiere magazine, I wrote “Starship Hollywood,” an attempt to outline the future of the movie biz.
I got a lot right, including home theaters, streaming, and consolidation of the big studios. But I really went to the edge in predicting that fake personas could stand in for real-life actors. What I didn’t predict was that I would write about this in a newsletter for WIRED magazine 31 years later.
When movies are fully digital it will be feasible to strip entire performances out of films, injecting new actors—either real people or computer-generated animations so lifelike that no one will be able to tell the difference. “Undoubtedly this will happen,” says Alexander Singer, who heads the Directors Guild committee on the future. “I think people will sell themselves for computer graphic simulations.
If Jack Nicholson is not available, he might license his persona for use. The price for licensing Jack Nicholson might be cheaper than the real Jack Nicholson, but not by much. Creation of characters from whole cloth is another matter.
” A high-tech casting director might digitize the body of Rita Hayworth, the lips of Madonna, the cheekbones of Nastassja Kinski, and Bette Davis’ eyes—animating the entire package to create the ultimate film goddess. If that isn’t ghoulish enough, consider the resurrection of popular but dead actors. Once the producer secured the licensing right, it would be simple, given unlimited computer power, to digitize the actors past performances and project them, with appropriate new animated actions, into original works.
Elvis could return to his annual film schedule. Michael asks, “Do you think Google’s watch can actually take on Apple Watch in a market so dominated by Apple?” Thanks for asking, Michael, and I appreciate that you are a repeat interlocutor. (Note to the rest of you—submit those questions! I’m running short!) You’re right that, after a bumpy start, the Apple Watch has risen to become the world’s most popular timepiece.
But the device draws much of its value from its tightly knit integration with the iPhone. The billions of people with Android could benefit from a ticker that works smoothly with their preferred operating system. And today’s announcement of the long-awaited Google Pixel Watch may tempt many of those non-iPhoners to at least take the time to consider it for their wrists.
Also, at $349 for the Wi-Fi version and only $399 for a cellular-equipped model, it’s cheaper than Apple Watch. There’s another reason I think that Apple’s dominance isn’t so inevitable. Smartwatches in general have yet to reach the Aristotelian perfection that we see in the smartphone and other mature products.
Apple’s most recent watch has added a lot of features but delivers them in a bulkier package. The perfect watch will not be packed with gimmicks or devoted to the needs of scuba divers, but will provide full-time, phone-free connectivity in a profile no fatter than a luxury watch. Of course, the company most likely to get there first will be Apple.
Sorry, Google. You can submit questions to mail@wired. com .
Write ASK LEVY in the subject line. Putin is still talking nukes . The head of Facebook explains why the Feed has moved from friend-connector to “discovery engine.
” Here’s our first impression of the Pixel Watch . Trump’s 2016 campaign rode to victory via Meta. The Dems are fighting back .
Here are the 100 hottest startups in Europe. .
From: wired
URL: https://www.wired.com/story/plaintext-bruce-willis-deepfake-metaverse/