OpenAI’s ChatGPT Christmas celebration is underway, with the AI company working its way through 12 days of announcements. Not all of them will be big reveals, though OpenAI has 12 daily livestreams scheduled for the event. With 10 left, we’re in for two weeks of ChatGPT-related novelties, and I’m sure one of them will be the public release of Sora.
Sora is OpenAI’s incredible text-to-video genAI product, which was announced earlier this year. We saw Sora in action several months ago, and OpenAI said it would release it to the public sometime in 2024. The company didn’t provide any release date information at the time, choosing to tread carefully in a year with crucial elections taking place worldwide. Sora, like other genAI programs, opens the door to potential abuse.
While we eagerly await Sora’s release, OpenAI is already working on a Sora 2 model that’s even more advanced. And now, a presentation including Sora 2 video demos has apparently leaked.
The Sora leak from late November was an indication that a public launch was close. OpenAI made it available to creators to get feedback on the text-to-video tool and fix issues ahead of a wider release. This tells me that OpenAI must be working on newer Sora features behind the scenes, including the so-called Sora 2 version that you’ll see below.
Tech. Entertainment. Science. Your inbox.
Sign up for the most interesting tech & entertainment news out there.
By signing up, I agree to the Terms of Use and have reviewed the Privacy Notice.
Sora 2 will probably be released at some point after Sora, though we have no idea when. Whatever the case, the next genAI tool is seemingly ready for some demos, like the one you’re going to see below.
It’s unclear where the demo comes from and whether it’s an internal or external presentation. I’d assume OpenAI demoed Sora 2 to third-party partners, one of whom filmed the segment above.
As you can see, the visuals continue to be stunning, and there appears to be some character permanence between different scenes. That will be a key feature for anyone trying to make longer AI movies. You’ll need to move the characters from scene to scene without any glitches. The current version of Sora isn’t very useful in that department
On the other hand, some character movements still look artificial in some scenes, so more work is probably needed to make the animation look like a real animation.
It’s not just the characters that are impressive in this short Sora 2 demo. Look at some of the objects, the prompts, and the scenery. OpenAI also gives us slow-motion, zooming, and other camera movement effects to suggest that Sora 2 can deliver results similar to what filmmakers would achieve while shooting real movies.
The Sora 2 demo ends with an OpenAI exec explaining the company’s advancements in rendering high-density pixel patterns while showing off a Sora 2 video of a Viking warrior woman, with a close-up of her face as the camera closes in on her.
The scene is part of a fight sequence, and the prompt is visible, albeit blurry, on the screen. The woman has mud and blood on her face, and Sora does a great job of keeping the patterns in place. it even accounts for the physics.
The Redditor who uploaded the clip only showed a small part of the Sora 2 demo. It’ll be interesting to see if OpenAI will showcase any of these capabilities when it launches the first version of Sora to the public. Again, I expect that to happen either this week or next.
