OpenAI has been promising to release its next-gen video generator model, Sora, since February. On Monday, the company finally dropped a working version of it as part of its “12 Days of OpenAI” event.
“This is a critical part of our AGI roadmap,” OpenAI CEO Sam Altman said during the company’s live stream.
According to the OpenAI team, Sora will be made available to Plus and Pro subscribers in the U.S. and around the world starting Monday afternoon.
YouTuber Marquis Brownlee reportedly got early access to the video generator and released a brief review on his channel on Monday morning. Sora appears to not be built atop GPT-4, as virtually all of OpenAI’s other generative tools are. The model is not available through the standard ChatGPT website, but instead through Sora.com (which is still not live as of this post’s publication).
The model is capable of generating videos in resolutions ranging from 480p to 1080p in lengths from 5 to 20 seconds, from either text prompts or reference images. It’s also capable of editing and extending existing video clips. ChatGPT Plus subscribers will be allowed up to 50 clip generations at up to 720p per month, and fewer videos at higher resolutions, each five seconds long. Pro users will be allowed unlimited generations at all resolutions and durations as long as 20 seconds. In addition to editing tools, Sora also offers a “storyboard” feature that will enable creators to combine multiple prompts into a single cinematic scene.
Brownlee notes that the model needs “a few minutes” to generate a 1080p clip, but notes “that’s also, like, right now, when almost no one else is using it. I kind of wonder how much longer it’ll take when this is just open for anyone to use.” Brownlee also points out that the model has significant difficulty in properly generating legs and their movements, with the front and rear legs swapping positions in unnatural and incomprehensible ways.
Unlike Grok 2, Sora will limit what its users can create and explicitly bars the generation of copyrighted subjects, people under the age of 18, and anything containing violence or “explicit themes.”
Despite OpenAI’s leading position in the AI industry, Sora has been beset by delays throughout its development, enabling competitors like Runway’s Gen-3 alpha, Kuaishou Technology’s Kling and Meta’s Movie Gen models to beat it to market. Sora was also recently (however briefly) publicly leaked by a group of beta testers, who accused the company of “art washing” the model’s capabilities.