- OpenAI statement added
- OpenAI has apparently blocked access to Sora for all early access users, and the model is no longer available on Hugging Face. The demo videos have also disappeared.
Update November 26, 2024:
OpenAI rejects the accusations from the artists' group. Hundreds of artists contributed to Sora's development through their participation in the Alpha program and helped prioritize new features and safeguards, the company says.
According to OpenAI, participation is voluntary, with no obligation to provide feedback or use the tool. Artists are only required to use the tool responsibly and not to disclose confidential details during the development phase.
"We’ve been excited to offer these artists free access and will continue supporting them through grants, events, and other programs," an OpenAI spokesperson said, confirming that Sora is still in "research preview."
Original article from November 26, 2024:
A group of "early testers, red teamers, and creative partners" has released OpenAI's Sora AI video model to protest what they say is unpaid development work.
The group says OpenAI, "a $150B valued company," asked hundreds of artists to test the system, find errors, and provide feedback without fair compensation. The only offered benefit was participation in a competition to showcase selected Sora-generated films.
According to the group, OpenAI offers "minimal compensation which pales in comparison to the substantial PR and marketing value OpenAI receives." The group also points out that OpenAI wants to approve all videos before they are posted.
Artists demand fair compensation
The group is protesting OpenAI's business practices by posting a version of Sora on Hugging Face, an AI model hosting site, under the username "PR-Puppets". According to the HF interface, the leaked model can create 10-second video clips at up to 1080p resolution.
The group emphasized they're not against AI art - after all, they joined OpenAI's program in the first place. Rather, they take issue with how OpenAI structured and ran the artist program.
In their letter, the group points to several open-source options that artists can use instead, such as CogVideoX, Mochi 1, LTX Video, and Pyramid Flow. They recognize that open-source tools often require technical expertise that many artists don't have. That's why, they argue, companies need to step in and make these tools more user-friendly - while ensuring artists get paid fairly for their contributions.
Several videos made with the leaked Sora model have appeared on the Hugging Face page, showing off its ability to create photorealistic footage.| Video: via Hugging Face
Due to high demand, the Hugging Face access is currently overloaded and will likely be removed soon. OpenAI hasn't commented on the leak.
Sora Turbo
When examining the leaked code, AI developer Tibor Blaho found that it taps directly into OpenAI's Sora API using fixed authentication data from Hugging Face. His analysis shows that Sora comes in several versions, with the leaked code likely running what OpenAI calls the "turbo" variant.
The code also shows that Sora includes different preset styles, including a "natural" mode, along with different modes of operation such as "simple compose". References to "video gen" in the API suggest that the system may have additional capabilities beyond what OpenAI has shown publicly.
While OpenAI first showed off Sora in February 2024, the company hasn't made it available to the public, reportedly only showing it to Hollywood and select groups. Recent reports suggest that a faster version may be coming soon—which could be the turbo version that appeared in this leak.
