OpenAI Launches Sora Video Generator With ‘Many Limitations’

OpenAI has finally launched Sora, its artificial intelligence model that creates videos based off of text. The company, led by CEO Sam Altman, warned on Monday that the new tool was released with “many limitations” that may affect the videos users create.

ChatGPT Plus subscribers, who pay $20 per month for the service, will now be able to create up to 50 videos at 480p resolution; those videos will last up to 5 seconds. And those who pay for the new ChatGPT Pro subscription, which runs $200 per month, will have “unlimited generations” and be able to create 20-second videos with 1080p resolution.

Beyond text prompts, Sora will also be able to create videos based on “images and video inputs.”

But for now, only a “subset” of users will be able to create new videos based on pictures and videos of real people, OpenAI clarified on Monday, due to the “potential for abuse.” Images of children will also not be allowed to generate new videos during this trial period, OpenAI said, as the company settles its “approach to safety.”

Here’s a look at a few of the examples OpenAI shared on Sora’s website; a screenshot of a video of bears roaming and a screenshot of a video of two old men in shiny hats:

OpenAI first introduced Sora back in February. It will be available to American users and “most countries internationally,” per the company’s YouTube livestream on Monday, but there is “no timetable” for its release in the U.K. or other parts of Europe.

“The version of Sora we are deploying has many limitations,” OpenAI said in its Monday blog post announcing Sora’s release. “It often generates unrealistic physics and struggles with complex actions over long durations. Although Sora Turbo is much faster than the February preview, we’re still working to make the technology affordable for everyone.”

In a review back in February, Bloomberg pointed out that Sora — like other text-to-video AI models — had some obvious constraints. They noted how Sora “struggles with stray body parts and a mixed understanding of physics.”

The company’s official entry into the text-to-video space comes with a fair amount of competition. Google released its Veo model last week, and Meta unveiled its text-to-video AI tool in October, just to name a few of the major tech companies in the space. Stability AI, which specializes in audio and visual content based on text prompts, also received a vote of confidence in September when James Cameron joined its board.

You can check the new tool out for yourself at Sora.com.