How to use Sora, OpenAI’s new video generating tool
What is Sora?
Sora is a powerful AI video generation model that can create videos from text prompts, animate images, or remix videos in new styles. OpenAI first previewed the model back in February, but today is the first time the company is releasing it for broader use.
What’s new about this release?
The core function of Sora—creating impressive videos with simple prompts—remains similar to what was previewed in February, but OpenAI worked to make the model faster and cheaper ahead of this wider release. There are a few new features, and two stand out.
One is called Storyboard. With it, you can create multiple AI-generated videos and then assemble them together on a timeline, much the way you would with conventional video editors like Adobe Premiere Pro.
The second is a feed that functions as a sort of creative gallery. Users can post their Sora-generated videos to the feed, see the prompts behind certain videos, tweak them, and generally get inspiration, OpenAI says.
How much can you do with it?
You can generate videos from text prompts, change the style of videos and change elements with a tool called Remix, and assemble multiple clips together with Storyboard. Sora also provides preset styles you can apply to your videos, like moody film noir or cardboard and papercraft, which gives a stop-motion feel. You can also trim and loop the videos that you make.
Who can use it?
To generate videos with Sora, you’ll need to subscribe to one of OpenAI’s premium plans—either ChatGPT Plus ($20 per month) or ChatGPT Pro ($200 per month). Both subscriptions include access to other OpenAI products as well. Users with ChatGPT Plus can generate videos as long as five seconds with a resolution up to 720p. This plan lets you create 50 videos per month.
Users with a ChatGPT Pro subscription can generate longer, higher-resolution videos, capped at a resolution of 1080p and a duration of 20 seconds. They can also have Sora generate up to five variations of a video at once from a single prompt, making it possible to review options faster. Pro users are limited to 500 videos per month but can also create unlimited “relaxed” videos, which are not generated in the moment but rather queued for when site traffic is low.
Both subscription levels make it possible to create videos in three aspect ratios: vertical, horizontal, and square.
If you don’t have a subscription, you’ll be limited to viewing the feed of Sora-generated videos.
OpenAI is starting its global launch of Sora today, but it will take longer to launch in “most of Europe,” the company said.
OPENAI
Where can I access it?
OpenAI has broken Sora out from ChatGPT. To access it, go to Sora.com and log in with your ChatGPT Plus or Pro account. (MIT Technology Review was unable to access the site at press time—a note on the site indicated that signups were paused because they were “currently experiencing heavy traffic.”)
How’d we get here?
A number of things have happened since OpenAI first unveiled Sora back in February. Other tech companies have also launched video generation tools, like Meta Movie Gen and Google Veo. There’s also been plenty of backlash. For example, artists who had early access to experiment with Sora leaked the tool to protest the way OpenAI has trained it on artists’ work without compensation.
What’s next?
As with any new release of a model, it remains to be seen what steps OpenAI has taken to keep Sora from being used for nefarious, illegal, or unethical purposes, like the creation of deepfakes. On the question of moderation and safety, an OpenAI employee said they “might not get it perfect on day one.”
Another looming question is how much computing capacity and energy Sora will use up every time it creates a video. Generating a video uses much more computing time, and therefore energy, than generating a typical text response in a tool like ChatGPT. The AI boom has already been an energy hog, presenting a challenge to tech companies aiming to rein in their emissions, and the wide availability of Sora and other video models like it has the potential to make that problem worse.
