The offices of Asteria Film are appropriately located in Los Angeles’ historic Mack Sennett Studios. Just as Canadian-born Sennett revolutionized motion pictures with his comedy shorts that drove audiences to early cinemas, film and animation studio Asteria and its parent, AI research company Moonvalley, are looking to shake up film and TV production with generative artificial intelligence (GAI).

“The story of Hollywood is a story of technology,” Bryn Mooser, founder of Asteria and co-founder of Moonvalley and non-fiction prodco XTR, tells Playback.

“Every time there’s a new technology that makes an image move, it brings in new filmmakers and studios. This building has witnessed a lot of change, including the transition from silent movies to sound. I hope people are reminded that the industry changes a lot, and [GAI] is just an evolution of that.”

Certainly, a sizable portion of the industry — from creatives to studio bosses — is enticed by the opportunities and potential cost efficiencies offered by GAI, which, for the most part, have been developed through working with existing content. But there’s also considerable controversy regarding training AI through the “scraping” of copyrighted IP. Lawsuits against GAI companies started in 2023, with the likes of Meta being sued for training its Llama AI model on pirated books.

Last year, Asteria partnered with Moonvalley to develop a foundational AI model dubbed Marey. (In another historical nod, it’s named after motion picture pioneer Étienne-Jules Marey.)

Moonvalley says Marey resolves the thorny copyright issue as an ethical – or “clean” – generative video model trained solely on data it has licensed. Launched this past March, Hollywood studios and filmmakers are currently experimenting with Marey, with specific partners to be announced in coming months as the product continues its roll-out.

Meanwhile, Asteria is putting Marey to work, building new workflows for animation and VFX that will compress the time it has traditionally taken to create or finesse shots.

It’s not a matter of simply typing a prompt and having Marey spit out a finished video. Filmmakers train a new AI model specific to their project on top of Marey, and then work in a process similar to creating CGI, but with the promise of five times greater efficiency.

For live-action content, it can simplify processes such as retargeting camera angles in post, extending shot duration, and replacing backgrounds. One project required epic Scandinavian landscapes, but didn’t have the location budget. So the producers mocked up a version of the exterior in studio, and are using Marey video-to-video transformation tools to transform it into the desired environment.

“In animation,” Mooser adds, “it can turn rough storyboard sketches into full-colour animatics or create new backgrounds or characters. We’re working on creating motion between frames.”

Ultimately what might be most exciting to the mid-level producer is the scale this potentially opens up.

“Currently, if you want to make a Pixar or Disney-type movie, you have to do it at a studio for US$150 million,” Mooser says. “But there’s a world where you can maybe make it for under US$10 million and finance it independently.”

The road to “responsible AI”

Marey is not at the point where it could be used to create, say, 90% of a production, but that might be only one generation away, according to Moonvalley CEO and co-founder Naeem Talukdar, who operates out of Guelph, Ont.

He says the content used to train Marey was licensed from indie prodcos, creative agencies, individual creators and even film students. The need for cleared footage has given rise to platforms such as Austin-headquartered Troveo, through which copyright holders can license content for availability to AI firms. Former CBC and TikTok exec Gave Lindo joined Troveo in November 2024 as its chief content officer, and leads the charge in liaising with rightsholders and content owners that want to further monetize their libraries through AI licensing.

Talukdar says Moonvalley has “gone as far as we can on the contracts to make sure the people we’re getting the data from have the rights to extend that data to us.” There would be legal ramifications if a finished project contained the unauthorized likeness of Cameron Diaz or a moon landing looking a little too much like a scene from First Man.

Indeed, in the U.S., those legal ramifications are being felt by tech firms such as OpenAI, which have been targeted with myriad lawsuits, including one launched in late November by a bevy of Canadian news outlets, such as CBC, The Globe and Mail and The Canadian Press. Meanwhile, Canada has seen the tabling of the Artificial Intelligence and Data Act, part of Bill C-27, which would legislate how businesses develop and use AI. The bill died when parliament was prorogued in January, and its future is up in the air.

“We’re excited by Canada being a potential driver in responsible AI and building the right regulatory frameworks that can become a model for other places,” says Talukdar. “At the same time, we need this to be paired with an optimistic approach. We want to avoid being so heavy on figuring out how to regulate AI that we’re not investing in its development.”

Larger studios can avoid these legal concerns by training AI on their own sizable libraries. This was a factor in a deal announced last September between the Canadian-American Lionsgate — which owns The Hunger Games movie franchise — and Runway AI, a New York tech firm specializing in products and models for creating GAI content.

“Having a catalogue of storied IP to start with puts you in a great position to begin exploring new stories within familiar worlds,” says Jamie Umpherson, head of Runway Studios (pictured right), the company’s creative arm.

Runway’s tools were featured in Oscar-winner Everything Everywhere All at Once (2022). In that film’s memorable interaction between two rocks, shot with green screen, VFX artist Evan Halleck told Variety that, using Runway automation, he cut out the characters and composited them on a plate shot in mere minutes — something that previously would have taken half a day.

Runway recently released its Gen-4 AI model, which it says solves one of video generation’s biggest problems: being able to generate specific people, places and things.

“You can generate consistent characters across endless lighting conditions, locations and treatments with a single reference image,” says Umpherson. “Or maybe you want to reshoot a scene from a different angle, but your actor isn’t able to come back on set. We’re making that easier.”

Details of the Runway/Lionsgate partnership remain under wraps, and Lionsgate did not respond to requests for comment for this story. But Umpherson says, “Our work with them goes back to the principles we’re focused on across the company — telling more stories and making more movies, all at a much lower cost, so that artists and studios can pursue novel ideas and give opportunities to new voices.”

Creative concerns

While Umpherson acknowledges that “generating story ideas and exploring new worlds are certainly part of this,” there has been strong pushback on how far AI should go in generating the fundamental creative elements of films and series.

In its newest agreement, the Writers Guild of America made sure Hollywood studios can’t generate scripts or edit a writer’s script with AI.

As well, the Writers Guild of Canada secured guarantees in its Independent Production Agreement that AI wouldn’t erode writers’ rights or have their content used for AI training.

Those concerned about the encroachment of GAI upon content production might be able to take some solace in the fact that even some of those on the AI vanguard feel the technology will never become the guiding creative force.

“Some tech companies say, ‘One day, anybody with a phone can become the next James Cameron,’ but no AI comes close to having taste, creativity and the essence of what it is to be a human artist,” says Moonvalley’s Talukdar. “And there’s no chance it’s going to be able to replace any of that.”

A version of this story originally appeared in Playback‘s Spring 2025 issue

Images courtesy of Runway Studio