In an industry prone to big emotional swings, artificial intelligence has produced a profound panic. Hollywood recently woke up to the news that fresh hell had arrived in the form of Sora 2, an OpenAI product that quickly and seamlessly creates videos with recognizable characters. Users can even insert themselves into the middle of the action—all for free, for now.
Within hours of the product’s launch, social media was awash in user-generated clips that injected figures from Star Wars, SpongeBob SquarePants, and Pokémon into various fantasies. This was Hollywood’s nightmare come to life: an AI technology that runs roughshod over the creative copyrights at the heart of the entertainment business. Sora offers “exploitation, not innovation,” warned the United Talent Agency, which represents artists and their precious intellectual property. Threats from Hollywood talent agencies, guilds, and other associations swiftly pushed OpenAI CEO Sam Altman to dial back Sora’s terms.
But the drama over Sora 2 stands in contrast with the slow progress AI technology has actually made in penetrating Hollywood. While the panic in the trenches is real, and the concerns over copyright infringement grab headlines—three major studios are suing the AI company Midjourney—AI has yet to yield the dreaded industry job losses. AI has yet to deliver on its promise to make filmmaking much cheaper and easier, too.
[Charlie Warzel: A tool that crushes creativity]
At a time when the entertainment industry is already in a slump, panic over the latest technological threat, or hope for some cost-saving technological salvation, makes sense. Eager to beat out the competition, big players are making costly bets on AI. Their sunk costs and hard-earned lessons are salutary for anyone who sees this whizzy new technology as either an industry-wide job killer or a lifeline for a shrinking business.
“Every director is playing with it in some way. Every actor is really scared of it. But from Hollywood’s point of view there’s still a misconception that AI is much more powerful and convincing than it really is,” Jed Weintrob, a founding partner at 30 Ninjas, an immersive-content studio that relies on AI, told me.
The Walt Disney Company, for example, made headlines last November with a big investment in a new division that will coordinate the company’s use of AI—a move that promised to harness the technology’s ability “to create some efficiencies,” as Disney’s CEO, Bob Iger, said last year about AI’s benefits more generally. Two industry executives I spoke with said that these savings have yet to materialize and that Iger is frustrated over sinking some $50 million into the effort. A Disney spokesperson acknowledged that getting the AI efficiencies off the ground with “technology that changes with every day has been a challenging one.” Ben Stanbury, a leading AI executive at Disney, quietly left this summer after less than a year in the role, the company confirmed to me.
It’s not that AI can’t do impressive things. It’s that folding disruptive technology into the kind of massive productions that Disney makes—such as Lilo & Stitch and Star Wars—is extremely difficult. For example, a postproduction visual-effects editor on a big movie usually has an airtight system for getting the job done. “There’s security, computing requirements. It’s all set within the studio. It’s a well-oiled machine,” Yves Bergquist, the director of the AI in Media project at the University of Southern California and an AI consultant for Hollywood studios, told me. “Suddenly you’re inserting third-party tools that are not part of Avid, Adobe, or any of these tools.”
Another problem studios have run into is that Hollywood legal departments don’t quite know whether AI-generated content can be protected by copyright. “It’s not clear if the level of human involvement involved provides copyright protection,” Bergquist said. “Legal departments and technical departments are clashing in every single studio, really and truly.”
Lionsgate’s pioneering partnership with the AI firm Runway—billed as the first deal between a major Hollywood studio and an AI company—has similarly fallen short of its stated goals. By training Runway’s AI on Lionsgate’s library, the studio hoped that the technology would generate “cinematic video,” storyboards for films and TV shows, and even spin-off adaptations of its most popular franchises. “Now we can say, ‘Do it in anime, make it PG-13.’ Three hours later, I’ll have the movie,” Lionsgate Vice Chairman Michael Burns predicted in an interview with New York magazine in June.
Yet many of these ambitions have gone unrealized. The company has no AI-generated films or adaptations in the works.
Lionsgate spokesperson Peter Wilkes told me that AI tools have been “very useful” in helping filmmakers with “previsualization and storyboarding,” and pointed to the upcoming Hunger Games prequel as an example in which the studio effectively harnessed AI. But he declined to give details about actual cost savings. Wilkes also denied that the plan was ever to create an entire film using AI—despite Burns’s suggestions otherwise.
A problem Lionsgate ran into, according to two executives familiar with the studio’s AI operations who asked not to be named for fear of retribution, is that AI needs to be trained on massive amounts of material in order to create anything credible. The library of one studio isn’t nearly enough.
More important, in order to create new iterations of existing movies or shows, the actors have to participate by giving AI models access to a fuller range of personal imagery—not just clips of characters that they’ve played. Actors, backed by their guilds and unions, are understandably uninterested in empowering a bot to do the work they are paid to do.
This is exactly what made the arrival of Tilly Norwood, an AI-generated “actor” who isn’t based on an existing person or intellectual property, so alarming. Norwood, a hyperreal ingenue created by Xicoia, an AI-talent studio, struck fear in the hearts of actors everywhere after its Zurich Film Festival debut last month. Xicoia’s founder, Eline Van Der Velden, announced at the festival that various talent agents had already expressed interest in representing Norwood for future projects.
Yet the Norwood short film was not very good, and Norwood was clearly limited in its ability to credibly represent humanity on-screen. Despite the flurry of concern from actors and SAG-AFTRA (the guild said it “believes creativity is, and should remain, human-centered”), there are no known plans for Norwood to become a hirable asset. “If she has a future, it won’t be at WME. We represent humans,” Richard Weitz, a co-chair of William Morris Endeavor—one of Hollywood’s two largest agencies—told me in September.
This isn’t to say that AI hasn’t infiltrated Hollywood. The tools are already being used to replicate animation and dub productions into multiple languages. Producers—mainly of low-budget films rather than major studio productions—told me that the technology is helping them reduce their spending on visual effects. Veteran directors including Darren Aronofsky and Doug Liman are experimenting with Google’s AI technology; James Cameron is working with Stability AI, a company whose board he sits on. Liman debuted a short film at the Venice Film Festival, Asteroid, that relied heavily on AI.
[Read: There’s no longer any doubt that Hollywood writing is powering AI]
But ultimately, as Tilly Norwood demonstrated and insiders affirmed, the AI models available just aren’t Hollywood-caliber—yet. “Hollywood studios have a very, very high bar of technical quality that AI currently doesn’t get. But it will,” Weintrob said. Studios may also be wary of a wholehearted embrace of AI owing to security concerns. “The high-performing models are made in China,” Bergquist said. “Do you want to download them on the company server?”
As a result, the tech remains a curiosity rather than a savvy investment or an existential threat—at least for now. There are even signs that AI and its uncanny promises are driving a greater appreciation for human craftsmanship.
At the recent premiere of Netflix’s gorgeous new adaptation of the Gothic horror classic Frankenstein, directed by Hollywood’s horror poet, Guillermo del Toro, Netflix’s co-CEO Ted Sarandos proudly touted the film’s lack of AI. He sounded wistful as he told me about visiting the set in Toronto, where a replica of a 19th-century ship had been built, down to the last detail.
Still, it would be hubris to assume that humans have won this round. Instead it seems more likely that films forgoing AI will soon be fetishized for their charmingly anachronistic human labor, like artisanal cheeses and handwoven textiles. This month, Netflix announced that it is merging its visual-effects studio, Scanline, with its research lab, Eyeline, to expedite its own AI-led efforts. The race to get ahead goes on.
