OpenAI’s Sora leaked by angry testers

Sora, OpenAI’s highly anticipated video-to-text model (first introduced in February), has been leaked to the public by a group of angry beta testers, who call themselves the ‘Sora PR Puppets’.

A video-to-text model, which was connected to a Sora API, was discovered on the public AI model repository platform, Hugging Face, allowing users to download and use the tool for the first time, despite it not being ready for the public.

With the model’s release on Hugging Face came a note slamming OpenAI for pressuring artists and testers on the Sora beta program—including red teamers and creative partners—to “provide unpaid labor through bug testing, feedback, and experimental work.”

They also berated OpenAI for misleading the public about Sora’s capabilities, as “every Sora output needs to be approved by OpenAI before it’s shared widely…and only a few creators in the program will be selected to have their Sora-created works screened.”

“We are not against the use of AI technology as a tool for the arts (if we were, we probably wouldn’t have been invited to this program). 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.”

Despite access to Sora being removed after just 3 hours, several users were able to use it to create 10-second clips based on text prompts, which they were quick to post to the social media platform, X. The feedback was mixed, with most saying although it was a faster “turbo” variant (it previously took 10 minutes to create a 60-second clip) that had multiple style controls, it did come with limited customization options, which is possibly a reason why it hasn’t yet been released to the public yet. That, and the numerous rumored technical setbacks that have been reportedly plaguing the model.