OpenAI Has Released Sora

OpenAI has officially rolled out its video generation model, Sora. Earlier this year, OpenAI unveiled the original version of Sora, which can generate realistic videos from text. This marks a significant leap forward in creating AI that can understand and simulate reality, ultimately setting the stage for models that can interact with the physical world.

OpenAI has developed a faster version of Sora called Sora Turbo. This new version launching today is available to ChatGPT Plus and Pro users as a standalone product. Compared to the version released in February, Sora Turbo is considerably faster, offering users the ability to generate higher-quality videos with impressive efficiency.

OpenAI has rolled out a wide range of features initially detailed in a technical report for those interested in the technical aspects. Now, users can generate videos with a resolution of up to 1080p, with each video having a maximum length of 20 seconds. Sora supports multiple aspect ratios, including widescreen, vertical, and square formats. Users can bring in their assets to remix, blend, or extend existing videos or create entirely new content from text.

New interfaces have been introduced to streamline interacting with Sora, making it easier for users to provide prompts through text, images, and videos. A helpful storyboard tool allows precise control over each frame, ensuring users can specify exactly what they want in their creations. In addition, Sora now constantly updates “Featured” and “Recent” feeds, showcasing the most recent and innovative works created by the community.

Regarding availability and subscriptions, Sora is included with the Plus plan at no extra cost, which allows users to generate up to 50 videos per month at 480p resolution or fewer at 720p. For those who need more, the Pro plan offers 10 times the usage, higher resolutions, and longer video durations. OpenAI also plans to offer tailored pricing options for different user types, which are expected to roll out early next year.

However, the current version of Sora still has its limitations. For instance, it sometimes struggles with rendering complex actions over long periods and can generate unrealistic physics. While Sora Turbo is much faster than the earlier preview, OpenAI is still working to make this technology more affordable and accessible to a broader audience.

OpenAI is releasing its video generation technology now to allow society to explore its possibilities and collaboratively develop norms and safeguards for its responsible use. Each video generated by Sora will come with C2PA metadata, marking the video as Sora-generated to promote transparency and facilitate origin verification. Although it’s still a work in progress, OpenAI has implemented safeguards, including visible watermarks and an internal search tool, to help verify whether Sora generated content.

OpenAI is taking a proactive approach to prevent harmful abuse, including blocking damaging content like child sexual abuse materials and deepfake videos. At launch, uploads of people’s likenesses will be restricted, but OpenAI plans to gradually lift these restrictions as it refines its deepfake mitigation strategies. Detailed information about safety measures and red teaming efforts is available in the system card.

With this early version of Sora, OpenAI hopes to empower people worldwide to explore new creative possibilities, tell their stories innovatively, and push the boundaries of video storytelling. The company is eager to see how users use this powerful tool and what kinds of groundbreaking videos the world will create using Sora.