Technology reporter Rowan Cheung revealed that a prototype of OpenAI’s unreleased Sora video generation model has been leaked online by a protest group calling themselves “Sora PR Puppets.”
According to Cheung, the leaked version of Sora seems to be more advanced than the February demo shown by OpenAI. Cheung noted that the leaked system is able to generate 1080p 10-second video clips at a much faster pace than the previously reported 10-minute render times.
Furthermore, the leaked Sora implementation appears to include new features like inpainting and image generation, which were rumored to be in development for a newer version of the model.
The leak has raised concerns from the protest group about OpenAI’s early access program for Sora. They claim the company recruited hundreds of artists to provide unpaid labor through extensive testing and feedback, all while maintaining strict control over the generated content.
The temporary Hugging Face instance hosting the leaked Sora model was quickly taken down, but not before users reportedly saw OpenAI’s watermark on the video clips.
This development comes several months after OpenAI indicated they were training an updated version of Sora to address technical challenges around video generation, though no confirmed release timeline had been provided.
OpenAI muted people on their official Discord server for discussing the Sora leak, which puts to rest any doubt that this leak may have been intentional.
In a statement, an OpenAI spokesperson said that Sora remains in a “research preview,” and that the company is “working to balance creativity with robust safety measures for broader use.” The spokesperson said “Hundreds of artists in our alpha have shaped Sora’s development, helping prioritize new features and safeguards,” and that “Participation is voluntary, with no obligation to provide feedback or use the tool.”
About the author:
