Runway CEO Says AI Could 'Absolutely' Replace Human Actors

Runway co-CEO Cristóbal Valenzuela told an audience at Turin's View Conference that AI could replace human actors. "Yeah, absolutely. I think it's already happening in a way," he said during a fireside chat titled "Every Pixel Will Be Generated," when an audience member asked whether AI could eventually take actors' place.

Valenzuela described AI as "a new medium of expression" and "a new medium of storytelling." His fellow speaker, Nvidia's Richard Kerris, pushed back on the timetable, if not the technology. Kerris, Nvidia VP of developer relations and general manager of media and entertainment, had earlier recalled a director's argument that the spontaneity between a director and an actor — Marlon Brando improvising with a cat on the set of "The Godfather," for instance — 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," Kerris said.

"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.

What does this change for filmmaking?

Kerris stressed that filmmaking still demands an understanding of character, composition and story. There is no "make movie button."

"I wish," joked Valenzuela.

The scope of the claim becomes clearer in Valenzuela's essay "It's All About the Pixel Economy," in which he 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."

A film today can take hundreds of people to make and gives millions of viewers two hours of watching. Valenzuela imagines audiences generating personalized entertainment as quickly as they consume it.

In a second essay, "The Cost of a Frame," Valenzuela charts the history of media through the cost of producing a credible frame. The layer that matters here is high-production-value content — the tier in which Hollywood, and many of the conference's professionals, operate. "This is the last expensive layer," he writes, and AI, he argues, is collapsing the cost of producing it.

How did Nvidia frame the resistance?

Kerris compared resistance to the technology with fears stirred by earlier advances, including 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, offered a similarly confident picture in his presentation, "Creative Control in the Age of AI Agents," claiming 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."

Not everyone at the conference shared that optimism. Between sessions, VFX artists, game studio heads and animation directors spoke less of fear than of anger — directed at something specific to generative models: the material used to train them.

Asked by Variety about the training data, 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."

That distinction may matter less to the courts, the likely arbiter. In July, a federal judge approved Anthropic's $1.5 billion settlement with authors whose books were downloaded from pirate libraries. A prior ruling 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. A proposed class action by visual artists against Stability AI, Midjourney, DeviantArt and Runway itself remains in the legal system, alleging their copyrighted work was used to develop generative models without compensation or permission.

Taken together, the talks from the companies leading AI at View Conference sketched a worldview in which the cost of generating moving images approaches zero and traditional production structures are reshaped — a prospect that now rests as much on pending court rulings as on model capability.