Sora Is Being Resurrected, But Only For Filmmakers This Time

Seven months after OpenAI killed Sora, its text-to-video model is about to rise from the dead. But the resurrection looks nothing like the original: instead of a viral consumer app, Sora may return as professional software for filmmakers, built and sold by a new startup with deep Hollywood roots.

According to a report from The Information published October 12, OpenAI is in talks with an unnamed startup that wants to take over Sora, improve it, and sell software or services built on it. The founders are a revealing trio: Jeffrey Katzenberg, the DreamWorks co-founder turned tech investor; Bill Peebles, who led Sora research at OpenAI; and Sujay Jaswa, a former chief financial officer at Dropbox. The three, who are associated with the investment firm WndrCo, founded the company months ago with AI video for filmmakers as its mission.

If a deal is reached, the startup would receive access deep enough to significantly modify the model, and OpenAI would take an equity stake in the company. Whether Sora's actual model weights, the parameter values that determine how it behaves, would transfer is still unclear. But given that Peebles led Sora's development, he already knows its architecture and training methods intimately, which makes the weights question less decisive than it sounds.

Why Sora died the first time

Sora was a genuine phenomenon in its first life. Previewed in February 2024, it launched as a standalone app in October 2025 and became the most-downloaded app in the iOS Photo and Video category within a day, with users generating viral clips featuring characters like Lara Croft and Mario. Two months later, Disney signed a three-year, $1 billion agreement to license more than 200 characters from Disney, Marvel, Pixar, and Star Wars for AI video generation, one of the largest commitments any entertainment company had made to generative AI.

Then, on March 24, 2026, OpenAI shut the whole thing down. The Sora app and an accompanying audio app were discontinued, and the Disney deal died with them. According to reporting from NBC News cited at the time, the shutdown was a strategic pivot: OpenAI was reallocating scarce compute resources toward text and code generation, which produce more revenue per unit of chip power, ahead of an expected IPO. The company had raised $110 billion at a reported valuation of around $730 billion, creating enormous pressure to focus on profitable services.

The warning signs were there. Peebles himself had announced usage limits on the Sora app months earlier, citing chip constraints. Executives told the Wall Street Journal the company simply could not do everything at once. Video generation, the most compute-hungry application OpenAI shipped, was the natural casualty.

The pros-only pivot

The new deal, if completed, would represent a fundamentally different bet: that AI video's future is not a consumer playground but a professional tool. The startup is explicitly building for filmmakers, not casual users. That reframing solves several of Sora's original problems at once.

First, economics. A consumer app needs massive scale to cover video generation's compute costs, and Sora never found a revenue model that worked. Professional software for film production can charge professional prices. Second, moderation. OpenAI struggled with deepfake abuse on the consumer app, a problem that shrinks when the user base is working studios with licenses and identities. Third, IP. The Disney collapse showed how fraught licensed characters were as a consumer feature. In a professional context, rights are negotiated production by production, the way the industry already works.

OpenAI could not make Sora pay as a toy for the internet. Handing it to filmmakers reframes it as an instrument, and instruments get priced like equipment, not entertainment.

The creative direction matters too. Sora's second-generation model produced videos with synchronized audio and improved physics, capabilities that map directly onto previs, shot planning, and rapid prototyping in production workflows. A filmmaker who can iterate on a scene's look before spending a day on a soundstage has a tool worth paying for. A user who can generate a ten-second Mario clip has a distraction worth burning chips on.

What it means for film culture

The symbolism of the founding team should not be overlooked. Katzenberg is not just a Hollywood name; he is a signal that the industry's old guard is now placing real bets on AI production tooling rather than fighting it from the outside. After years of strikes, lawsuits, and the collapse of the Disney experiment, this deal treats AI video as settled infrastructure: the question is no longer whether it enters filmmaking, but who controls the tools and who gets paid.

That is the sharper edge of the story. If Sora becomes a licensed tool used in professional production, the industry's talent contracts and licensing frameworks can wrap around it. The guilds negotiated AI protections precisely for this scenario: human performers credited, paid, and kept in the loop while machines handle the intermediate work. A pro-only Sora fits inside that framework far more cleanly than a consumer app where anyone could synthesize anyone.

There are open questions, and honest coverage should name them. The deal is not done, and the terms of the equity stake and the weights question are unknown. OpenAI has not confirmed anything publicly, and The Information's report rests on a single familiar source. Google's Veo 3, ByteDance's tools, and Runway have all moved ahead in the consumer and prosumer space Sora vacated. A filmmaker-focused Sora would enter a market that is no longer waiting for it.

Still, the trajectory is instructive. The history of creative technology is full of tools that failed as consumer products and succeeded as professional ones: desktop publishing software, non-linear editing, even the synthesizer. Sora's second life may prove the pattern again. The model that was too expensive to entertain the internet might turn out to be exactly expensive enough to matter to the people who make what the internet watches.