News Scrap

Key Takeaways

A group leaked OpenAI’s Sora video tool through a frontend on Hugging Face.
The group accuses OpenAI of exploiting testers and misusing their unpaid work.

OpenAI’s long-awaited video generation model, Sora, has leaked, according to a TechCrunch report.

A group appears to have leaked access to the video generation model through Hugging Face, an AI developer platform, which is reportedly connected to OpenAI’s Sora API, not yet publicly available.

The Hugging Face repository, dubbed “PR Puppet Sora,” allows users to generate 10-second videos in 1080p resolution through a frontend linked to the API.

The frontend, accessible to users, includes a section titled “Open Letter: Why Are We Doing This,” where the group states that the leak is a protest against OpenAI’s alleged exploitation of unpaid labor and strict narrative control in its early access program.

Frontend Manifesto: Group protest against OpenAI

“Hundreds of artists provide unpaid labor through bug testing, feedback and experimental work for the [Sora early access] program for a $150B valued company,” the group stated in a post accompanying the frontend.

The group claims that the early access program prioritizes public relations and advertisement over creative expression and critique.

According to the group, OpenAI requires approval before any Sora-generated content can be shared, and only select creators will have their works screened.

“We are not against the use of AI technology as a tool for the arts,” they wrote. “What we don’t agree with is how this artist program has been rolled out and how the tool is shaping up ahead of a possible public release.”

The leaked version appears to be a faster “turbo” variant of Sora, based on code discovered by users on X. The original system, unveiled in February, required over 10 minutes to generate a 1-minute video clip, according to the report.

The frontend has recently been updated to state that OpenAI has temporarily shut down Sora’s early access program for all artists.

Many videos have surfaced in the past few hours following the news, but numerous users on X claim that the model is not as good as expected, despite OpenAI “gatekeeping” it for months.

User “Chubby” on X shared videos reportedly generated with Sora’s leaked model, depicting movements and aesthetics similar to previously released text-to-image models.

This has led many to conclude that Sora’s model does not live up to the high expectations, despite OpenAI delaying its release to “perfect the model, get safety/impersonation/other things right, and scale compute,” according to Chief Product Officer Kevin Weil.

The company has also faced additional challenges, including the departure of Sora co-lead Tim Brooks to Google in October, while competitors like Runway and Stability AI have gained ground by securing partnerships with major film studios.

Updated with new information from OpenAI restricting access to the Hugging Face frontend