OPENAI
Sora 2 might be the craziest AI tool we’ve ever seen
OpenAI has released Sora 2, its new video-and-audio generation model.
The first Sora came out in early 2024 and showed the first signs that AI could really handle video.
Now, Sora 2 is a huge step forward.
Earlier models often broke reality to follow prompts - objects would stretch or teleport, for example.
Sora 2 is better at following real-world rules: balls bounce off backboards, paddleboards float properly, and even mistakes look more natural.
It can also manage complex instructions, keep track of scenes across multiple shots, and add realistic sound, speech, and effects.
As usual, the examples uploaded by OpenAI themselves are mind-blowing. Check them out.
People can even “upload” themselves, placing their likeness and voice inside generated videos.
To showcase this feature, OpenAI has launched a social iOS app called Sora.
It lets users create, remix, and share videos.
Yep, we got the AI version of Vine.
Here’s Sora 2 in brief:
More realistic video: Physics and mistakes look natural.
New Sora app: Social platform built around cameos.
Safety built in: Parental tools, consent controls, and wellbeing checks.
Invite-only but chaos inside
The key feature is cameos, where friends can drop each other into scenes.
The app is designed to encourage making, not endless scrolling, and comes with tools to manage feeds, parental controls, and safeguards around likeness and safety.
Sora 2 is first rolling out in the US and Canada, with wider access coming in the coming weeks.
It’s free for now with some limits, and a Pro version is available for ChatGPT Pro users. An API release is also coming.
Sora 2 makes my iMovie edits look like cave paintings. And that’s okay.