(Bloomberg) -- OpenAI is rolling out an artificial intelligence system called Sora that can generate realistic-looking videos from text prompts, nearly 10 months after the startup first publicly previewed the technology.
Sora will be available to paid users of chatbot ChatGPT in the US and other markets on Monday, the company said in a livestreamed presentation. The new version of the tool will be able to churn out videos up to 20 seconds in length and offer multiple variations of those clips.
With Sora, OpenAI is working to keep pace with a growing number of startups that now offer tools to spit out short clips quickly and more cheaply than creating video from scratch — potentially upending the livelihoods of many in Hollywood. The list of rival AI video companies includes Pika, Haiper and Runway, the last of which recently partnered with a studio to train a new AI model and explore how the technology can be used in film production.
After OpenAI teased Sora in February, the ChatGPT maker began courting film studios, media executives and talent agencies in a series of meetings. OpenAI also granted access to a number of artists, as well as a few big-name actors and directors, to try out the service.
Early on, Sora had clear constraints, however. In an early test, Bloomberg News found Sora struggled with stray body parts and a mixed understanding of physics. It also required noticeably more time — and computing power — to generate each video than users might be accustomed to with an image generator or chatbot.
Initially, at least, Sora’s availability will still be limited. Those who pay $20 per month for a ChatGPT Plus subscription will be able to generate 50 videos monthly, said OpenAI Chief Executive Officer Sam Altman. Users who pay $200 per month for a ChatGPT Pro subscription will have unlimited access to “slower” generations, he said, and 500 videos created at a faster speed.
To start, Sora also won’t be available in most of Europe, which has tougher tech regulations.
During the demonstration Monday, OpenAI showed off Sora’s new features. The tool can produce videos from still images and videos as well as text prompts, the company said. For instance, a user could upload a video clip and the software will spit out footage to lengthen it.
Sora also includes an option called Storyboard, which allows people to write out what they want to happen at different points in a video to better direct a sequence of clips. And there’s an option to explore a feed of videos created by other users.
This week’s announcement came as part of what OpenAI has said will be 12 days of livestreamed product events. On Thursday, OpenAI kicked off the launch series by introducing the more expensive new ChatGPT Pro option, with unlimited access to its most powerful AI models.
(Adds details on new version of Sora starting in second paragraph.)
©2024 Bloomberg L.P.