In early 2024, Stability AI—once hailed as a pioneer in generative artificial intelligence—was seconds away from insolvency when two tech veterans staged a dramatic intervention. Today, the company aims to outflank rivals by pivoting squarely toward Hollywood.
At a candlelit gathering in February 2024, Lady Gaga and Sean Parker welcomed friends into her $22.5 million Malibu greenhouse to celebrate the launch of a skin-care nonprofit. Parker, the Napster co-founder and Facebook’s first president, mingled with Gaga’s guests as a Grammy-winning string quartet played. Executive-style focaccia and charcoal-grilled branzino circulated beneath floor-to-ceiling windows overlooking the Pacific.
Among the invitees was Prem Akkaraju, a longtime friend and business partner of Parker. The two had crossed paths when Parker was rising at Facebook and Akkaraju worked in the music industry. They’d failed to break into streaming, then bought a renowned visual-effects studio together. Lately, whispers of an AI venture had surfaced in their conversations.
That night, an investor in Stability AI leaned toward Akkaraju and warned that the company was “circling the drain”—just days from running out of options. “You should take Stability and make it into the Hollywood-friendly AI model,” the investor urged.
Hollywood, after all, was in a slump. Since 2022, US film and television production had plunged roughly 40 percent amid soaring budgets, international competition and protracted labor strikes by the Writers Guild of America and SAG-AFTRA. Studios and streamers raced to harness AI for tasks from automated dialogue translation to frame-by-frame visual effects, hoping to rein in costs and accelerate schedules. Startups such as Luma, Runway and Asteria were pitching tools to blockbusters and indies alike.
Akkaraju saw an opening. Stability AI already boasted the breakthrough text-to-image model Stable Diffusion, released in mid-2022. All it needed, he believed, was a Hollywood gloss and steady leadership. There was just one hitch: the company already had a founder-CEO.
Emad Mostaque, a former hedge-fund manager, launched Stability AI in 2020 on a mission to “build systems that make a real difference” solving social challenges. By 2022, he had assembled—or rented—a cloud supercomputer to host generative models. Spotting OpenAI’s closed-source success, he sought an open-source counterpart “like Linux to Windows.” He offered compute time to academics developing a system that turned text prompts into images. In August 2022, they unveiled Stable Diffusion with Stability AI’s backing.
Stable Diffusion exploded, drawing 10 million users in its first two months. “It was fairly close to state-of-the-art,” says Maneesh Agrawala, a Stanford University computer science professor. Because it was open source, researchers could tweak and extend it, spawning a vibrant community. With just 77 employees by that October, Stability AI rode the wave of thousands of community contributors. Mostaque then secured $101 million in seed funding from Coatue, Lightspeed and others—achieving unicorn status almost overnight.
Inside the company, the atmosphere was electric. “It was an incredibly fun and chaotic startup that was throwing a lot of spaghetti at the wall, and some of it stuck really hard,” a former high-ranking employee recalls on condition of anonymity. Mostaque envisioned everything from drug discovery to “Game of Thrones” scripts generated by AI.
But as excitement soared, management woes emerged. “On the research side, we did really good things. The other side I was not so good at, which was the management side,” Mostaque admits now. Two former staffers say he focused on models, not on forging a sustainable product roadmap.
At the same time, legal and reputational challenges mounted. Stable Diffusion 1.5 had been trained on LAION-5B, a dataset scraped from 5.8 billion web images—some containing child-abuse content and copyrighted photographs. In January 2023, Getty Images sued Stability AI in London’s High Court for training its model on 12 million Getty photos without permission, branding it “brazen theft and freeriding.” A parallel suit followed in the United States.
Then, in June 2023, Forbes published a report accusing Mostaque of inflating his résumé and misrepresenting Stability’s partnerships. He had claimed a partnership with Amazon Web Services when in fact it was a standard cloud-services deal. He also overstated his Oxford credentials—though he insists the confusion stemmed from a clerical error.
Confidence evaporated. Four months after the exposé, investors from Coatue and Lightspeed resigned from the board. By year’s end, the head of research, COO, general counsel and HR chief had all departed, followed by many prominent researchers. Under investor pressure, Mostaque left on March 22, 2024—just weeks after that greenhouse soirée.
Within days, Parker and Akkaraju assumed control, naming Akkaraju CEO and Parker chairman of the board. They never consulted Mostaque directly, though he says he offered help.
Their first order of business: retool Stability AI for the film industry’s exacting demands. But rivals were already surging. In September 2023, Runway struck the sector’s landmark deal: Lionsgate granted it access to its entire film catalog as training data. “The time it takes to go from idea to execution is just shrinking,” says Runway CEO Cristóbal Valenzuela. “You can do things in minutes that used to take weeks.” He forecasts small teams replacing massive crews and multimillion-dollar budgets.
The Lionsgate agreement accelerated AI adoption. “Last year it was ‘let’s prototype,’” observes Amit Jain, CEO of competitor Luma. “This year, it’s a whole different tone.” Moonvalley, founded by former DeepMind researchers and parent to Natasha Lyonne’s AI film studio Asteria, says over a dozen major studios are trialing its latest model.
Akkaraju credits success to “the three T’s: timing, team and technology.” In his $20 million Beverly Hills home, he reclines on a white couch overlooking manicured gardens, his tailored shirt straining over sculpted arms. He delivers eye contact and handshakes with equal force.
Early in his tenure, he decided Stability would cede the frontier-model race to OpenAI and Google. Instead, it would package apps that plug into those giants’ systems, slashing Stability’s compute expenses. A renegotiated cloud contract wiped out its AWS debt. He won back investors without disclosing specifics.
Today, Stability sells itself as a software-as-a-service provider, not a model-builder. “Our differentiation is having the creator in the center,” Akkaraju says. He notes that no other AI firm counts James Cameron on its board.
The Cameron hire feels surreal. The director of “The Terminator” had once derided Parker’s Screening Room streaming platform—meant to let viewers watch new theatrical releases at home—calling himself “committed to the theater experience.” By 2020, the venture rebranded as SR Labs after studios rebuffed it. That same year, Parker and Akkaraju acquired Weta Digital, the New Zealand effects house behind “The Lord of the Rings” and Cameron’s “Avatar” films.
One night over dinner—with tequilas flowing—Cameron, Parker and Akkaraju found common ground. “A friendship formed,” Cameron says. “It was very funny.” Akkaraju recalls never formally apologizing; mutual respect did the rest.
Parker emphasizes Stability’s embrace of open source and respect for intellectual property. “That sounds rich,” he laughs, “given Napster and early social media. But it is a lesson learned.”
In June 2024, Getty dropped its UK claims as trial neared verdict, though the US case continues. Akkaraju stresses that Stability “sources data from publicly available and licensed datasets for training and fine-tuning,” adding that client projects are tuned exclusively on customer-provided material. “That’s the majority of what we’re using, for sure,” he says.
Yet generative AI remains imperfect for final-cut VFX. A filmmaker who asked to remain unnamed tried using a text-to-image tool for a Netflix shot, only to have it rejected for failing 4K quality checks. Consistency is another sticking point: feed the same prompt into most AI video or image systems ten times, and you’ll get ten different results. “We need higher resolution, higher repeatability, controllability at levels that aren’t quite there yet,” Cameron agrees.
Still, directors and producers rely on AI for previsualization. “The inefficiency in the old system was the information gap between what I see and what I imagine,” says Luisa Huang, co-founder of Toonstar. “With AI, the inefficiency becomes, ‘Here’s a version, here’s another version, here’s another version.’”
Jon Irwin, who directs Amazon’s biblical series House of David, discovered AI on location in Greece. He watched his production designer conjure concepts instantly. “I was like, ‘Tell me what you’re using, magician,’” Irwin recalls. He pitched Amazon on using generative AI to augment practical shoots. “We film everything we can for real—it still takes hundreds of people,” he says. “But we do it at about a third of the budget and twice the speed.” A burning-forest sequence that would have been prohibitively expensive in reality was fully generated by AI.
Irwin has spoken with Stability’s team but “not been able to use their tools successfully on a show at scale.” Many filmmakers have experimented with Stability’s models, yet none have fully integrated them into major productions.
Meanwhile, Netflix co-CEO Ted Sarandos told investors in July that the streamer used “gen AI final footage” in an original series, boosting production speed tenfold and slashing costs. “AI represents an incredible opportunity to help creators make films and series better, not just cheaper,” he said.
Back at Stability’s Los Angeles offices, chief technology officer Hanno Basse demos a model that turns a single backyard photo into a navigable 3D space. The software estimates depth, fills missing details and lets users choose camera moves—zoom, pan or spiral—from a menu. “Instead of spending days building a virtual environment, you take one image and generate a concept,” Basse explains.
Rob Legato, the company’s chief pipeline architect and an Oscar-winning VFX veteran, sits in as both executive and beta tester. He praises the vision but critiques the interface: “You probably want sliders instead of a dropdown.” And while the virtual camera tool shows promise, Legato cautions it isn’t yet film-ready. He’s equally bullish on AI-driven rotoscoping—automating the painstaking frame-by-frame masking process—but warns that artists still fear for their jobs. “If you guys are going to lose your jobs, you’re going to lose your jobs over work drying up versus getting bumped aside by these gen AI models,” Cameron says.
Akkaraju and Parker insist that cheaper productions will create more films, expanding employment. Akkaraju invokes the ATM metaphor: tellers worried that machines would replace them, yet bank-teller jobs and pay grew after ATMs rolled out. That argument, once a staple of techno-optimists like Eric Schmidt, obscures the fact that teller numbers peaked in 2015 and have since declined.
Whether Stability AI’s comeback will outlast its rocky past remains to be seen, but one thing is clear: generative AI is already reshaping how movies and shows are made. From a near-collapse in a Malibu greenhouse to a bid for Hollywood’s heart, the story of Stability AI captures both the promise and the peril of this next wave in entertainment technology.