It is now almost impossible to browse the web without being bombarded with AI-generated videos. Open up any social media platform and you’ll see a clip of an uncanny-looking fake natural disaster, or animals doing impossible tasks. The majority of videos are terrible. They’re almost always accompanied with hundreds, if no thousands, of comments and likes from people who insist that AI-generated contents are a new artform that’s going change the world.
This is especially true for AI clips that are intended to look realistic. It doesn’t matter how bizarre or inconsistent the footage is, someone will always say that the entertainment industry needs to be concerned. The idea that AI generated video is both the future and a threat to Hollywood has spread like wildfire among supporters of the relatively new technology.
It’s hard to imagine major studios adopting this technology in its current form, when the output of AI models is often not good enough to be turned into a movie or TV series. Bryn Mooser, a filmmaker who launched Asteria last year, wants to change that impression. He’s also partnering with Natasha Lyonne, a producer at Late Night Labsa studio that focuses on generative AI, which Mooser acquired last year. Asteria’s biggest selling point
is that unlike most AI outfits, its generative model, which it built with research firm Moonvalley has been “ethical,” which means it has only trained on properly licensed materials. The concept of ethical generative AI could become a key part of the way AI is adopted in the entertainment industry, especially after Disney and Universal sued Midjourney for copyright violations. Mooser told me in a recent conversation that Asteria’s understanding of what generative artificial intelligence is and isn’t, helps it stand out from other players within the AI space.
Mooser says, “As filmmakers, we were aware that AI was presented in Hollywood in a way that was not very effective. It was clear that the tools were not being built by anyone who had ever made a movie before. Text-to-video, where you can say’make a new Star Wars film’ and it will happen, was something that Silicon Valley believed people wanted and could actually do. People have been able to reproduce singers’ voices using generative AIwith ease and produce passable music. Mooser believes that in the rush to normalize gen-AI, the tech industry has conflated audio with visual output, which is at odds with what makes good films.
Mooser says, “You can’t tell Christopher Nolan to use this tool and text your own way to The Odyssey ()” As Hollywood users began to use these tools, it became clear that the form factor was not going to work. The amount of control a filmmaker requires is often at the pixel-level. It would allow, for example, an artist to create a model which could generate assets in their unique style and then use it populate a whole world of characters and objects adhering to a unique aesthetic. Asteria used this workflow to produce the animated short “A Love Letter To LA” by musician Cuco. By training Asteria’s model on 60 original drawings drawn by artist Paul Flores (19459012), the studio was able to generate new 2D assets, and convert them to 3D models that were used to build the video’s fictional town. The short is impressive but its heavy style speaks to how projects that use generative AI as their core often must work within the visual limitations of the technology. This workflow doesn’t seem to offer control down to a pixel level yet.
Mooser states that, depending on the agreement between Asteria’s clients and Asteria, filmmakers may retain partial ownership of models after they are completed. Asteria is also “exploring” revenue sharing options in addition to the licensing fees it pays the creators of material that its core model is based on. Mooser’s focus is on wooing artists with the promise of lower production and development costs.
Mooser says, “If you are doing a Pixar animation film, you may be coming on as a writer or director, but you won’t often have any ownership in what you’re creating, residuals or a cut of what they make when they sell a box,” he tells me. “But if you use this technology to reduce the cost and make it independently financing, then you can have a world that allows real ownership to be possible.” The live-action movie is about a teenager whose shaky sense of reality leads her to see the world in a video game-like way. Asteria will create many of the Uncanny Valley’sfantastical visual elements that are similar to Matrix. This detail makes Uncanny Valleyseem like a project that is designed to present the hallucinatory anomalies that generative AI is known for as clever features instead of bugs. Mooser told me that he hoped “nobody thought about the AI part at all” since “everything will have the director’s touch on it.” “That was very obvious as we thought about this. I don’t believe anyone wants to see what computers create.
Mooser, like many generative AI supporters, sees the technology a “democratizing tool” that can make art creation more accessible. He also emphasizes that, under the correct circumstances, generative artificial intelligence could make it easier for a movie to be produced for around $10-20 millions instead of $150 million. Yet, securing this kind of capital can be a challenge for many younger, up-and coming filmmakers.
Mooser has repeatedly mentioned to me that generative AI can produce finished works quicker and with smaller teams. He described this aspect of an AI workflow as a positive, which would allow writers and director to work more closely and efficiently with key collaborators such as art and VFX Supervisors without having to spend so much money on revisions. By definition, smaller teams mean fewer jobs. This raises the question of whether AI could put people out to work. When I bring up this issue with Mooser he points out the recent closure by Technicolor Group of the VFX house as an example of how the entertainment industry has been undergoing a constant upheaval, which began to leave workers unemployed even before the generative AI hype reached its current fever pitch.
Mooser did not want to minimize the fact that Hollywood’s double strike in 2023 was a result of these concerns about generative artificial intelligence. He is adamant that many workers in the industry will be able pivot laterally to new careers based on generative AI, if they embrace the technology.
Mooser says that there are filmmakers and VFX artist who are adaptable, and want to embrace this moment in the same way as people were able switch from editing film to editing with Avid. This technology offers a great opportunity for people who are true technicians, such as art directors, cinematographers and writers. It’s important that we, as an industry, know what’s good and bad about this technology, what’s helpful to us in telling our stories, and, most importantly, what’s dangerous. Asteria’s workflows are one of its biggest selling points. However, if they became industry standard, it would be difficult to scale them in a manner that could accommodate the entertainment workforce as it transitions into new careers. Mooser is a master at highlighting the positive aspects of his product. Now he must prove that his technology — and the changes it brings — can work.
