Durng a fireside chat entitled “Every Pixel Will Be Generated” at Turin’s View Conference, an audience member asked Cristóbal Valenzuela, the co-CEO of generative AI startup Runway, whether AI could eventually replace human actors.
“Yeah, absolutely. I think it’s already happening in a way,” he replied, going on to describe AI as “a new medium of expression” and “a new medium of storytelling.”
His fellow speaker, Nvidia‘s Richard Kerris, was somewhat more attached to existing arrangements. Earlier in the session, he had recalled a director’s argument that the spontaneity between a director and an actor, such as Marlon Brando improvising with a cat on the set of “The Godfather,” could never be prompted.
“That craft is something to behold, and I just think that that’s something that I don’t think we’ll ever lose,” said Kerris, Nvidia VP of developer relations and general manager of media and entertainment.
“Not now, but I think models are going to get probably to a point where you can probably direct them in a very similar way,” Valenzuela replied.
Kerris had also stressed that filmmaking still requires an understanding of character, composition and story. There is no “make movie button.”
“I wish,” joked Valenzuela.
A clear sense of what this might mean sits in Vaenzuela’s essay titled “It’s All About the Pixel Economy,” in which Valenzuela predicts that “the value of creating pixels is trending towards zero.” Eventually, he writes, the number of “seconds consumed” will converge with the number of “seconds generated.”
Presently, a film often takes hundreds of people to make and millions of people two hours to watch. Valenzuela imagines a future where audiences generate personalized entertainment as quickly as they consume it.
In another essay, “The Cost of a Frame,” Valenzuela charts a history of media through the cost of producing a credible frame. The relevant layer is the high production value content. It’s that tier in which Hollywood, and much of the conference’s professionals, work in.
“This is the last expensive layer,” he writes and AI, he makes the case, is collapsing the cost of producing it.
At the View fireside chat, Kerris frequently framed the resistance toward the technology as no different than the fears that previous tech advances stirred, including the emergence of computer-generated imagery. “The similarities of fear and rush to judgment on things is so similar to what I’ve seen in the past,” he said.
Jeff Kember, Nvidia’s principal technical evangelist for media, painted a similarly optimistic vision in his presentation, “Creative Control in the Age of AI Agents,” claiming that adoption is already widespread.
“There’s a whole bunch of people talking about how they’re using generative AI and then there’s also a bunch of people who aren’t talking about it and they’re not talking about it primarily because their legal team is not letting them talk about it,” he said. ‘But both people on each side of this fence are both using generative AI. I don’t know anyone who isn’t.”
Between sessions a number of VFX artists, game studio heads, and animation directors spoke less of fear and more of anger relating to something peculiar and unique to the generative models, namely the material used to train on in the first place.
Asked by Variety about the training data behind the models, Valenzuela said one of the “core misbeliefs” about AI was that “people thought AI was some sort of a database.”
“They’re not databases,” he said. “They’re learning from representation.”
It seems an unavoidable concern with the courts, the likely arbiter. In July, a federal judge approved Anthropic’s settlement of $1.5 billion with authors whose books were downloaded from pirate libraries. A prior ruling had found that using copyrighted works to train models could constitute fair use, while distinguishing that question from the company’s acquisition and retention of pirated copies. Meanwhile, there are a number of lawsuits against many of the biggest AI companies still going through the legal system. One is a proposed class action case by visual artists against Stability AI, Midjourney, DeviantArt and Runway itself. It alleges that their copyrighted work was used to develop the generative models without compensation or permission.
Taken together the talks from the companies leading the charge in AI at View Conference revealed a worldview whereby the cost of generating moving images approaches zero and at the least the traditional structures of production are fundamentally reshaped.