Though moviegoers certainly have an appetite for franchise slop these days, it’s a stretch to say that they’d be willing to bite down on wholly AI-generated movies just yet.
Even more of a stretch: that the technology is anywhere near good enough to produce entire feature films.
This is the reckoning that the mammoth Hollywood studio Lionsgate is undergoing right now, as one of the first big studios to go in on the tech. Almost exactly a year ago, it announced a bold partnership with the AI startup Runway to develop a new model capable of generating “cinematic video” exclusively for Lionsgate to use. In return, the studio gave the firm unrestricted access to its treasure trove of movies — which include everything from the “Hunger Games” films to “American Psycho” — to train the AI model.
In an interview with Vulture last month, Lionsgate vice chairman Michael Burns gloated about the AI model’s possibilities, like being able to churn out rehashed versions of stuff from the studio’s catalog. “Now we can say, ‘Do it in anime,'” Burns said — and out comes an animated version of “John Wick.”
It sounds great in theory. But in practice, it’s been a slowly unfolding disaster, reports The Wrap, with copyright woes and the tech’s fundamental shortcomings coming to bear on development. More damningly, even an entire studio’s worth of movies as training data hasn’t been enough to make an AI model that’s worthy of ripping them off.
“The Lionsgate catalog is too small to create a model,” a source told The Wrap. “In fact, the Disney catalog is too small to create a model.”
Disney has had its own struggles with incorporating the tech, including a failed stab at AI-deepfaking Dwayne “The Rock” Johnson’s face into an upcoming live-action “Moana” sequel. After a laborious 18 months of collaborating with the firm Metaphysic to create Johnson’s AI clone, none of the AI-assisted scenes they shot will make it into the final film.
The faults largely lie with the tech itself. A shortage in quality training data has been plaguing the broader AI industry, and it’s raised questions of whether the current method of improving AI models is sustainable, if it hasn’t already hit a wall. If scraping the entire internet for training footage wasn’t enough to make a refined AI video tool — and then further fine-tuning it on a Hollywood studio’s catalog — what will be?
Adding to the woes, who owns or gets to do what with a movie in Hollywood is infamously Byzantine. A lot of people are involved in producing them, and there’s usually multiple parties in charge of distributing the finished product.
The entire modus operandi of the generative AI industry, meanwhile, straddles incredibly gray legal territory. Many top AI models were trained without getting permission to use the data vacuumed up online, the legal consequences of which are now taking shape. Recently, leading AI firm Anthropic settled with authors to pay them $1.5 billion for feeding their books into its models. And at this point, it’s not even clear that wholly AI-generated content is even copyrightable, according to Jacob Noti-Victor, a professor at Cardozo Law School.
... continue reading