Navigating the internet today often involves encountering AI-generated videos, a seemingly ubiquitous presence on social media platforms. Users are frequently met with bizarre clips showcasing fabricated natural disasters or animals performing remarkable feats. Despite their often subpar quality, these videos often attract extensive engagement, with many users lauding AI-generated content as a groundbreaking new art form poised to transform the creative landscape.
This trend is particularly evident in AI clips crafted to resemble realistic footage. Regardless of their oddities or disjointed aesthetics, supporters often voice concerns that the rise of AI-generated videos could cast a shadow over the traditional entertainment industry. Advocates of this emerging technology assert that AI’s potential in video production represents both an exciting future for filmmaking and a significant challenge for Hollywood.
However, the prospect of major studios integrating such technology raises questions, particularly given that the current output from AI models often lacks the quality necessary for mainstream films or series. Filmmaker Bryn Mooser aims to alter this perception through Asteria, a production company he founded last year, alongside a new AI-generated feature film being developed by Natasha Lyonne, who is also his partner and advisor at Late Night Labs, a studio dedicated to generative AI.
Asteria differentiates itself from other AI companies by promoting its generative model as “ethical,” developed in collaboration with Moonvalley and trained exclusively on properly licensed content. Following recent legal actions against AI tools for copyright infringement from major studios including Disney and Universal, the notion of ethical generative AI is expected to gain traction across the entertainment sector. Mooser emphasizes that Asteria’s nuanced understanding of generative AI distinguishes it from others in the field.
“When we began to conceptualize Asteria, it became evident that there were significant problems with how AI was marketed to Hollywood,” says Mooser. “The tools currently in development often lack input from actual filmmakers. The idea that one could simply command a creation like ‘make me a new Star Wars movie’ reflects a misunderstanding of the filmmaking process prevalent in Silicon Valley.”
Mooser believes that the perception of generative video models as threats to traditional filmmaking is rooted in a misconception that AI can seamlessly replicate real-world footage, similar to how AI can imitate musical styles. While the technology has demonstrated a capacity for replicating voices and creating music, Mooser argues that conflating audio and visual outputs overlooks the elements that contribute to quality storytelling in cinema.
“One cannot simply request a tool to create a masterpiece akin to The Odyssey,” he states. “As filmmakers began exploring these tools, it quickly became apparent that the underlying format was insufficient. True control for filmmakers often requires manipulation at a pixel level.”
To enhance creative control for filmmakers, Asteria utilizes its primary model, Marey, to create project-specific models trained on original visual material. This approach allows artists to generate assets reflective of their unique styles, facilitating the creation of coherent worlds populated by distinct characters and objects. An example of this workflow is seen in Asteria’s production of musician Cuco’s animated short, “A Love Letter to LA.” By training its model on 60 original illustrations by artist Paul Flores, Asteria generated new 2D assets and converted them into 3D models to construct a fictional town for the video. Although impressive, the project highlights the stylistic constraints that often accompany generative AI, emphasizing that achieving pixel-level control is still a work in progress.
Mooser indicates that, depending on financial terms agreed upon with clients, filmmakers may retain partial ownership of the AI models once completed. Alongside traditional licensing fees paid to creators whose material informs Asteria’s core model, the studio is also exploring revenue-sharing frameworks. For now, however, Mooser aims to persuade artists by highlighting the potential for lower initial development and production costs.
“In traditional Pixar productions, you might either direct or write but rarely gain any lasting ownership of the project or its revenue,” he explains. “By leveraging this technology to reduce costs, we can create a financing model that enables real ownership for creators.”
Asteria will put its beliefs about the transformative potential of generative AI to the test with Uncanny Valley, a feature film co-written and directed by Lyonne. The film follows a teenage girl whose uncertain grasp of reality leads her to perceive her surroundings as increasingly video game-like. Many of the film’s visually striking elements are set to be developed with Asteria’s proprietary models. This choice suggests an effort to depict the surreal qualities associated with generative AI in a manner that serves the narrative rather than detracts from it. Mooser hopes audiences will focus on the artistic vision rather than the technology used in its creation.
“It’s not a matter of simply instructing, ‘then they enter a video game,’ to see what unfolds, as that is not engaging for viewers,” he asserts. “This is a significant consideration as we conceptualized the film. The audience’s interest lies in more than just the algorithms.”
Mooser shares the view of many proponents of generative AI, believing that it has the potential to democratize art creation. He emphasizes that, under optimal circumstances, filmmakers could produce a film for a fraction of the standard budget, signaling a shift toward more accessible filmmaking. Nevertheless, acquiring such funding remains a significant hurdle for many emerging filmmakers.
One of Asteria’s notable advantages, as highlighted by Mooser, is the ability of generative AI to expedite the production process while minimizing team size. He describes this aspect as beneficial, allowing writers and directors to collaborate closely with essential team members such as art and VFX supervisors without the delays often caused by revisions involving larger groups. However, smaller teams inherently raise concerns about job security, prompting discussion regarding the potential for AI to displace human workers. Mooser cites the recent closure of Technicolor Group as an instance of the industry’s ongoing transition, which began to leave professionals without work before the current generative AI surge.
Mooser recognizes that worries about generative AI play a significant role in the industry upheaval that led to the double strike in Hollywood in 2023. However, he maintains that many entertainment professionals can adapt to new roles that incorporate generative AI if they are willing to embrace this technology.
“There are filmmakers and VFX specialists who are flexible and eager to embrace this transformation, similar to the shift from film editing to Avid,” he notes. “Technicians across various roles—art directors, cinematographers, writers, directors, and actors—have a chance to leverage this technology. It is crucial for our industry to discern what aspects of generative AI are beneficial for storytelling and which could pose dangers.”
The potential risks associated with Hollywood’s growing fascination with generative AI lie not in the obsolescence of the studio system but in the technology’s capacity to enable studios to function with fewer human resources. This reality is a core aspect of Asteria’s model, leading to concerns about how the industry can accommodate a shifting workforce. While Mooser articulates the advantages of generative AI, the challenge now lies in demonstrating that his approach—and the transformations it entails—can be successful.
