OpenAI’s Sora text-to-video tool leaked by aggrieved beta testers

A group of artists and beta testers have leaked OpenAI’s text-to-video tool Sora after they claim to have been exploited in its research and development phase.
Tom Mitchelhill
Posted by Tom Mitchelhill November 27, 2024 2 Mins Read
A group of artists and beta testers have leaked OpenAI’s text-to-video tool Sora after they claim to have been exploited in its research and development phase.

A group of aggrieved artists and early testers of OpenAI’s unreleased text-to-video tool Sora leaked access to the new model in an act of protest after they claimed to have been used for “unpaid research and development.”

The group published what appears to be a front-end version of Sora to AI developer platform HuggingFace on Nov. 26, allowing anyone to utilize the tool. OpenAI has reportedly since intervened to shut it down.

Technology, OpenAI

Artists and beta testers claim OpenAI exploited them in its latest text-to-video model Sora. Source: Hugging Face

The leak was carried out by a group of artists and beta testers operating collectively under the username “PR-Puppets.” 

“We received access to Sora with the promise to be early testers, red teamers, and creative partners. However, we believe that instead, we are being lured into ‘art washing’ to tell the world that Sora is a useful tool for artists,” the group said.

According to the open letter published alongside the leak of Sora, the group claimed that “hundreds of artists” provided unpaid labor through bug testing, feedback and experimental work only to find themselves excluded from any compensation or recognition from OpenAI.

The artists claim that OpenAI — now privately valued at $157 billion — was unfairly preventing artists and contributors from being paid for testing, feedback and development. 

The tool was online for several hours before it was taken down, and a number of users were quick to post examples of videos generated by the leaked version of the tool on X. 

“It’s impressive how well it handles arms and legs,” wrote film director Huang Lu in a Nov. 26 post to X, alongside a clip from the leaked tool.

The leaked version of Sora appears to be a faster “turbo” variant, according to code uncovered by X users, with additional lines of code also hinting at certain controls on customization and styles of video to be generated in the future. 

Related: Salesforce CEO says LLM ‘upper limits’ reached, future of AI is agents

Sora was first unveiled by OpenAI on Feb. 16 and wowed X users with the level of hyper-realistic video content it could generate from simple prompts. 

According to a Feb. 17 report from The Information, OpenAI had been training Sora on “hundreds of millions of hours” of video clips in a bid to cover a wide range of styles and improve the overall quality of its AI-generated footage. 

OpenAI did not immediately respond to a request for comment.

Magazine: Crypto has 4 years to grow so big ‘no one can shut it down’ — Kain Warwick, Infinex

Tags: