A leak concerning OpenAI’s unreleased video generation model, Sora, surfaced on Tuesday, revealing early access granted to a group protesting the company’s approaches to “art washing,” as noted by X user @legit_rumors.
This group, known as Sora PR Puppets, obtained the Sora API access and utilized authentication tokens to create a user-friendly interface, allowing anyone to generate video clips. Although the initiative was shut down within three hours—likely by Hugging Face or OpenAI—many participants posted their creations on various social media platforms.
Check it out here: https://t.co/gnnkoj0jc2
If Sora seems like an optimized version. Can produce up to 1080p 10-second clips.
Consider duplicating the space (my attempt was unsuccessful!).
One example: pic.twitter.com/npphRJgyrd— Kol Tregaskes (@koltregaskes) November 26, 2024
“Hundreds of artists contribute unpaid efforts through bug testing, feedback, and experimental work for a company valued at $150 billion,” the Sora PR Puppets outlined in their manifesto. “This early access initiative seems to focus more on public relations and advertising rather than fostering genuine creative expression and critique.”
The group also criticized OpenAI for vetting videos created during the early access period, implying that the company required approval before such clips could be shared. “By releasing this tool, we provide everyone the chance to experiment with what approximately 300 artists were offered: unrestricted and free access to this tool,” they stated.
Confirmed: OpenAI Sora has indeed been leaked https://t.co/Vh1zzsKgPT pic.twitter.com/mAN1Z4vGsN
— Chubby♨️ (@kimmonismus) November 26, 2024
Another example from the public pic.twitter.com/nVPw1rZVB5
— ʟᴇɢɪᴛ (@legit_rumors) November 26, 2024
Users who managed to access the interface before it was removed were able to generate short video clips up to ten seconds long in 1080p resolution. Reports indicate that this version is a streamlined iteration of what OpenAI revealed earlier this year.
Previously, that early version of Sora reportedly had a sluggish processing speed, requiring up to ten minutes of real-world time to create a one-minute video and often failed to maintain a consistent aesthetic throughout.
As OpenAI faces increasing competition in video generation technology, its adversaries are catching up rapidly. Competitors like Meta’s Movie Gen and Kuaishou’s Kling not only rival Sora’s quality but have entered the market, while OpenAI continues to develop its offering.
In a recent Reddit AMA, OpenAI’s chief product officer, Kevin Weil, noted that the release of Sora is being postponed due to the need for further refinement of the model, ensuring safety, impersonation issues, and scaling computing capabilities.
