Hanno Basse, chief technology officer at AI tech developer Stability AI—whose board members include filmmaker James Cameron—spoke about his company’s work and the evolution of generative AI in filmmaking during this week’s HPA Tech Retreat. “The ultimate goal is to make more movies,” he stated.
Reflecting on the years it can take to complete a Hollywood production, such as Cameron’s Avatar films, he added, “We heard this from a number of other filmmakers as well: ‘We wanted to make more films in the past, and the time it takes for us, that’s the problem.’”
Speaking at the annual event in Palm Springs, where AI was a major topic, Basse addressed issues such as data provenance, production budgets, and advancements in filmmaking. He predicted that within “probably two years, we’re going to have very high-performing open-source foundation models.”
Stability AI specializes in generative AI models for various media, including image, video, 3D, audio, and language, with its flagship model being Stable Diffusion.
“Our guiding principles, basically, are that generative AI needs to be part of the production process and not replace it,” Basse explained. He emphasized the company’s “artist-centric” and “task-centric” approaches, highlighting recent innovations such as 3D tools and a prototype for aspect ratio conversion.
The company also features prominent Hollywood figures in leadership roles. In addition to Cameron—whom Basse described as “really actively involved”—the board chairman is entrepreneur and Napster co-founder Sean Parker, while the CEO is Prem Akkaraju, the former head of VFX powerhouse Weta. Basse himself previously served as CTO at Fox before Disney’s acquisition, as well as at VFX studio Digital Domain.
Stability AI and Azure
During his talk, Basse offered insights into AI’s role in the industry. “I don’t know anything that’s impossible in the industry today, especially with the advancements of visual effects.
So I don’t think [AI] is about the impossible,” he said. “It is really about making things faster and cheaper. If you look at your budgets today, whether it’s TV or film, 87-90% of that budget is human labor. And I’m not talking about putting people out of work, but there’s a lot of that labor that is not as fun to do.”
Text-to-video technology was another focal point at the retreat, though Basse downplayed its significance, suggesting it’s a passing “fad” due to filmmakers’ need for maximum control. “
When it actually comes to producing a two-hour movie with high production value that a lot of people will want to go and see, I think we’re very far from that in terms of generative AI being able to do that,” he asserted.
Basse also addressed concerns about data provenance, predicting that it will become less of an issue over time. “I actually think that, going forward, the provenance issue is going to be less of a problem than we all think it is now,” he said, explaining that production will require “very specific” data.
“We’re not going to get that data from scraping YouTube or whatever. We’re going to get that data only from people like [Hollywood professionals], and we’re already negotiating training data deals with people. And there’s term sheets; there’s very clear conditions as to what this data can be used for.”
He concluded by affirming Stability AI’s commitment to working within industry standards: “We are here to play ball, and to play by the rules of the industry.”
