Ten months after first tease, OpenAI launches Sora video generation publicly
Briefly

OpenAI showcased the "Explore" interface allowing users to browse video prompts for free, while video generation requires a subscription. Additionally, the new "Storyboard" feature enables frame-by-frame direction.
OpenAI released Sora's System Card detailing the technical aspects of its model, including how it uses visual patches, compressing videos into a latent space for effective representation.
Sora employs a recaptioning technique similar to DALL-E 3, generating descriptive captions for visual data that allows more accurate adherence to user text instructions in generated videos.
Safety measures include C2PA metadata in generated videos for identification and visible watermarks to ensure content authenticity.
Read at Ars Technica
[
|
]