Today, OpenAI launched Sora 2. The latest generative AI text-to-video model comes after the first launch of Sora back in February 2024, before finally being released at scale in December 2024 just under a year ago.
Sora 2’s performance is seriously impressive, showing a significant leap forward in realism. The videos produced with Sora 2 aren’t just visually stunning, the model has improved dramatically in terms of its understanding of real-world physics. This means when characters and objects interact with their environments, they do a much better job of reflecting reality, like gravity, or the limits of how the human body moves and bends.
The term uncanny valley describes the feeling of unease that occurs when a robot or computer-generated character is almost, but not quite, human. I’ve experienced this a number of times today as users shared their creations using Sora 2.
The videos are close, perhaps as close as we’ve ever seen to creating content that is so lifelike, so close to reality that it’s genuinely hard to tell if it’s real or not, but every now and then, you’ll spot something that feels off. The good news is, Sora 2 will be followed by Sora 3 and Sora 4 in the years to come and it’s abundantly clear that the day where you can’t tell, is rapidly approaching.
Today’s launch came with a launch video that begins with OpenAI CEO Sam Altman speaking. Having watched a number of intereviews with Sam, I know his physical appearance and I know his voice and the video was close, really close, but there were tiny differences that revealled this was an AI version of Altman.
This is the first time we get to see a new feature known as a ‘Cameo’. This allows you to go through a quick face scanning process with your phone to create a digital representation of yourself. Once you have a Cameo, you get to prompt against it, asking your digital self to appear in videos which can be really powerful.
OpenAI were clearly conscious of the potential concerns with this feature and spoke to the control users have over their Cameo. As you create a Cameo in the new Sora 2 mobile app, you get to decide if it’s shared with friends, the public, or nobody at all. If you have friends that share with you, you can then prompt the model to create videos of you and them together.
The Sora 2 app that was announced and shown today, is effectively a TikTok clone (and very similar to Instagram Reels, and YouTube Shorts), with one massive difference – every video on there is created with AI. This raises the question, do people want to watch AI-generated content. On other video platforms, there’s a mix of normal (aka real video) and some AI generated, that should be labelled as such.
The videos were really impressive, from skateboarding, to tennis, product commercials, dancing competition, movie trainers and comedy, these all integrate digital humans into their AI-powered videos. If you’re a content creator, or a tv and movie producer, Sora 2 should definitely have your attention.
Sora 2 is capable of generating high-definition (not 4k yet) and video clips of up to 16 seconds, complete with synchronised audio, dialogue, and sound effects. This supasses Google’s Veo 3 which previously was the best video generation model and could create 8 second clips.
The copyright question
Another major point of contention is copyright. OpenAI has stated that Sora 2 will use copyrighted content in its video generation unless the rights holders explicitly opt out.
This has already led to some high-profile companies, such as Disney, opting out of having their content used on the platform. The legal and ethical implications of this approach are sure to be a hot topic of debate in the months to come.
“I think they are certainly opening themselves up to lawsuits in a particular cases,”
Mark Lemley, a professor at Stanford Law School.
What the internet is saying
The online reaction to Sora 2 has been explosive. X has been flooded with examples of what the new model is capable of. Some of the most impressive examples showcase the model’s ability to create stunningly realistic and imaginative scenes.
One user, Reza Wrecktion, shared an incredible video of a futuristic cityscape, complete with flying vehicles and neon-lit buildings.
It’s been tough keeping this quiet…@OpenAI Sora 2 + the Sora App are here!
Super fun, super addicting—
An uncanny valley killer. What’s real?
Now I can share full commercials, trailers & more w/ sound, all from one prompt.
Texts are old news—
soon everyone will send… pic.twitter.com/HTPSFxu89n
— Reza Sixo Safai (@rezawrecktion) September 30, 2025
OpenAI’s own X account has also been sharing impressive demos, including one of CEO Sam Altman and researcher Bill Peebles on the moon, riding a jet ski in the Arctic, and even on the back of a dragon.
How to get access
For now, access to Sora 2 is limited to an invite-only beta in the US and Canada. There is no word yet on when it will be available in Australia. You can download the Sora app from the iOS App Store and sign up to be notified when it becomes available in your region.
How much does it cost?
The good news is that Sora 2 will initially be free to use, with what OpenAI describes as “generous limits.” There will also be a “Sora 2 Pro” version available to ChatGPT Pro subscribers, which will offer higher quality video generation.
Sora 2 is a glimpse into a future where anyone can be a video creator, limited only by their imagination. One thing is for sure: the world of video will never be the same.
For more information, head to https://openai.com/index/sora-2/
