The relationship between Hollywood and the AI industry is extremely difficult. Many filmmakers and industry workers dislike the idea of using AI in the Moviemaking process and worry that it will be used as a pale alternative to human creativity. In June, Disney and Universal sued AI company Midjourney for copyright infringement.
But nonetheless, some AI companies are trying to push their way forward in Hollywood. Also, on Tuesday, one such startup, Moonvalley, took a big step by bringing the fully licensed professional video model to the public.
Moonvalley was founded by Deepmind researchers and has a close relationship with the film industry. The company owns AI Film Studio Asteria Film Co. It was co-founded this year by filmmaker and actress Natasha Lyonne and filmmaker Bryn Mooser. Asteria advises Moonvalley on the development of the AI model Marey. This is now available to filmmakers for subscription tiers of $14.99, $34.99 and $149.99 per month.
Marey is developed with filmmaker approval and trained in licensed data, which could become a major entry point for AI, and in theory avoids the ethical issues and copyright lawsuits that studios plague the AI industry.
“We need to build these tools the right way. Rather than trying to automate work with filmmakers and artists, we need to build at the heart of that,” says Naeem Talukdar, CEO and co-founder of Moonvalley, on time.
Moonvalley has raised more than $100 million from investors, including Khosla Ventures and Bessemer Venture Partners. Asteria uses Marie in her new documentary about Carl Sagan to restore and adjust the footage. Talukdar also says that Marey is being tested in pilot programs not only in major advertising companies but also in more than 12 “large studios.”
Many other AI video models are black box systems. Enter the prompt to generate a wholesale of the scene. When you try to fine-tune one variable in your scene, another variable can change, making it difficult to maintain control over everything that was filmed.
Moonvalley aims to build tools that integrate into the filmmaking process, as CGI and special effects programs have done in the past. Marey gives filmmakers hypothetically that they allow filmmakers to enter storyboards and frames and then fine-tune them to fit, giving them a much more control over every detail, from objects to characters, motion, motion, scene composition.
“It's this iterative process that starts with input guidance and builds towards the scene you want. This isn't too different to how VFX workflows do today,” says Talukdar. “If you're an independent studio that doesn't necessarily have a large infrastructure, even small spaces can create and curate these scenes in great detail.”
Talukdar walked me through Marey's demonstrations, first showing actual video footage of the actress in the studio, turning her head and calmly pulling the gun from under her scarf. Tarkudar then ran the footage through a model to create another AI-generated woman scene on a train that moved just like the original actress. He then changed the camera view, spinned to her other side, and appeared the mountain outside the train window. Each scene generated costs the creator about $1 to $2. This is comparable to the cost of other AI video generators and is much cheaper than physically re-shooting the footage.
All footage on which the model is trained is licensed from the IP owner. About 80% of that footage comes from groups of independent filmmakers and institutions who have accumulated B-rolls over the years, says Talukdar. This approach means that Marey is trained as a competitor with about a fifth of data, like Google's VEO 3, Talukdar says. However, he argues that Moonvalley is overcoming this shortage with better technology created by graduates of Deepmind, Meta and other top labs.
“The reality is if we cut it off [data]our model will definitely be stronger,” he says. “But our trend is that you don't necessarily have to be the number one model. And I think this is the first generative, completely licensed model that doesn't have to compromise on quality.”
There are many filmmakers in Hollywood who consider AI to be the opposite of their creative process. The tension played a major role during the 2023 Hollywood strike, with many expressing fears about unemployment through automation. Conversely, Talukdar argues that AI tools will actually create new types of jobs and allow studios to push their budgets even further, rather than cutting them down.
“There's the idea that instead of spending $50 million on a movie, you can do that for $5 million. There are some truths in it,” he says. “But another way to think about it is the way every studio we were talking about is thinking about it — it's now $50 million and for the same 100 people on that project, they could do something that previously cost $100 million,” he says. “It's the same number of people and they're doing better content.”
Marie's supporters in the film industry include Anchel Manuel Soto. Blue beetles Other movies. “I feel like Moon Valley and Asteria have heard artists' concerns about ethical AI. What they created at Marie is a breakthrough,” he wrote in an email on time. “From streamlined studio workflows to empowering emerging creators in places like Puerto Rico or Dakar, Marie is the first generative AI to get what she really needs. A way to protect people who move faster, take responsibility for less, and make this industry human.”
However, many filmmakers have cast skepticism about this logic and are afraid that they are set up for bait and switch. “When you look at the bigger applications of these technologies, companies and studios don't want artists to be able to use it to make cooler things with the same amount of money.” Bojack Horsemantold Brookings last year. “They want to make things cheaper, cut out artists, pay people and use these technologies in ways that aren't even better.”
