10
0 Comments

OpenAI launches Sora, its text-to-video AI model
IH+ Subscribers Only

OpenAI has officially launched Sora, its text-to-video AI model, for ChatGPT Plus and Pro subscribers.

A screenshot of an AI-generated video with the following words overlayed: "OpenAI Sora"
  • OpenAI launches Sora, its text-to-video AI model.

  • Pro and Plus subscribers can access a "Turbo" version at sora.com

  • Sora is not yet available in most European countries.

Sora, which lets users generated videos up to 20 seconds long, is now available for all ChatGPT Plus and ChatGPT Pro subscribers.

That is, all subscribers except those in the EU. A new supported countries page on OpenAI's website shows that other excluded regions include Switzerland and the UK.

The model is a "Turbo" version of one OpenAI previewed back in February. It's much faster and can create videos at 1080p quality.

How long it takes to generate a video varies, but some users have reported creating 10-second clips in less than a minute.

Users create videos through a new ChatGPT-style interface, where they can upload their own assets to incorporate into clips, or ask Sora to make everything. Right now, Sora can produce videos in widescreen, vertical and square aspect ratios.

Is Sora any good?

Example clips on Sora.com show the model can create everything from planetary flybys to wrestling bears. The colors and details are vivid, but the motion on some of the videos is a little unnatural.

Like the image generation models that came before it, Sora still seems to struggle creating lifelike human hands.

The hands of a woman moving stones on a table start failing at around the 12-second mark on a video generated by X user Alex Patrascu. The hands themselves, however, are “spectacular.”

Patrascu, who posted several videos to X, praised Sora’s “Blend” feature, which can create smooth transitions between vastly different shots.

For longer videos, the model usually produced a series of quick, choppy shots. But occasionally, it generated a long shot that slowly shifted in focus, he added.

“Uncanny valley not quite crossed yet, but we're getting there...😅” wrote X user cloudzombi.

Others questioned if the model would ever be capable of creating longer videos with more complex narratives.

“Will we ever see a video more than 5 seconds that coherently tells a story” asked user bearish92. 

For what it’s worth, OpenAI doesn’t exactly position Sora as an aspiring filmmaker. The company describes Sora as:

 “A foundation for AI that understands and simulates reality — an important step towards developing models that can interact with the physical world.”

Lower quality for Plus users

Sora is currently available to ChatGPT Pro and ChatGPT Plus users in most countries outside of Europe. Some non-EU countries like Iceland can access Sora, but others like the U.K. and Norway cannot. OpenAI says it’s working to “expand access further in the coming months.”

At the moment, Plus users can only generate short videos at up to 720p quality. Longer and higher quality videos (and 10x as much usage) are only available to users on OpenAI’s new Pro plan, which costs $200 a month.

Plus, which provides up to 50 480p videos per month, or fewer at 720p quality, costs $20 per month.

The roll-out hasn't been totally smooth so far, with some users reporting capacity issues on X. Some say they've got stuck in generation request queues, while others have encountered "Sign ups are temporarily unavailable" messages.

Other Sora limitations

OpenAI admits that this version of Sora has a whole host of technical limitations. “It often generates unrealistic physics and struggles with complex actions over long durations,” the company stated.

It's also placed intentional restrictions on Sora intentionally to try and make it harder for users to generate harmful material.

To prevent users generating deepfakes, uploads of images and clips involving people will be limited for the time being. Sexual content is also blocked to prevent abuse.

Videos will also include a watermark and special metadata that shows they were generated by Sora.

OpenAI said: “We’re introducing our video generation technology now to give society time to explore its possibilities and co-develop norms and safeguards that ensure it’s used responsibly as the field advances.” 

Photo of Channing Allen Channing Allen

Channing Allen is the co-founder of Indie Hackers, where he helps share the stories, business ideas, strategies, and revenue numbers from the founders of profitable online businesses. Originally started in 2016, Indie Hackers would go on to be acquired by Stripe in 2017. Then in 2023, Channing and his co-founder spun Indie Hackers out of Stripe to return to their roots as a truly indie business.

Photo of Katie Hignett Katie Hignett

Katie is a journalist for Indie Hackers who specializes in tech, startups, exclusive investigations, and breaking news. She's written for Forbes, Newsweek, and more. She's also an indie hacker herself, working on EasyFOI.

Create a free account
to read this article.

Already have an account? Sign in.