Sora 2 launched Tuesday, alongside a brand new iOS app that's basically “TikTok meets AI video creation meets your face on anything.” And honestly? It's wild in ways both terrifying and really fun.
The app works more like a creation tool than a doomscroll machine. You can remix other people's videos, browse a customizable feed powered by LLM-based recommendation algorithms (yes, you can literally instruct the algorithm in natural language), and most importantly, it's designed to maximize active creation over passive consumption.
What Sora 2 can do
- Generate Olympic-level gymnastics routines, backflips on paddleboards (with accurate buoyancy!), and triple axels.
- Follow intricate multi-shot instructions while maintaining world state across scenes.
- Create realistic background soundscapes, dialogue, and sound effects automatically.
- Insert YOU into any video after a quick one-time recording (they call this "cameos").
The best video to show what it can do is probably this one, from OpenAI researcher Gabriel Peters, that depicts the behind the scenes of Sora 2 launch day…
As usual, Matt Berman provided a great recap on the news. Check out his video here:
That last feature, Cameos, is the big deal.
The new Sora app (iOS only for now) lets you record yourself once, then drop into any genAI scene with accurate appearance and voice. OpenAI tested it internally last week and people started making new friends at the company because of it. It looks fun enough for me to actually consider uploading my likeness, despite the very real and concerning privacy implications of that…
Oh that last point: OpenAI went HARD on safety here. Every video includes visible watermarks plus C2PA metadata. Only you control your cameo: you decide who can use it (only you, friends, people you approve, or the general public) and can revoke access anytime (you can even delete content someone else makes). There are stricter limits for teens, parental controls via ChatGPT, and the feed won't show teen profiles to adults.
Overcoming the creeped out feeling of uploading yourself to an AI is probably the #1 barrier to this reaching mass adoption, so we’ll see if their techniques are enough to convince folks they’ve handled enough risks that it’s worth trying it out for the lolz.
Who gets access and when:
Rolling out in the US and Canada first via invite-based system (sign up in the app for a notification). Initially free with generous limits to all users. Also, ChatGPT Pro users get access to the higher-quality Sora 2 Pro model here. The API is coming soon, and Sora 1 Turbo stays available for anyone who needs it.
Our Take
Gymnastics has recently become the new benchmark for video AI: think full routines with twists, flips, and rotations that follow actual biomechanics.
Kling 2.5 Turbo has been absolutely crushing this lately, generating sequences that track momentum, body position, and angular physics across multiple moves. It's become the unofficial stress test for whether your video model actually gets how bodies move through space.
So how does Sora 2 compare? In OpenAI's launch demo, they showed an ice skater doing a triple axel with a cat sitting on their head. The physics looked shockingly legit (weight distribution, momentum, rotational dynamics), even though realistically that cat would've gone flying after half a second. It's the kind of thing that makes robotics researchers salivate because these models are becoming true simulators of reality.
As we wrote on Monday, between their new “anti”-social network Pulse and now Sora 2, OpenAI seems to be building better social networks. Unlike other platforms that spiral us into addictive swipers, OpenAI's betting on giving us control over our algorithms (you can choose if you want relaxing content, or “animals only”, etc) and turning us all into active creators and remixers. We'll see if that actually works in practice... either way, Sora 2 looks like a ton of fun, and at a level of “realistic” quality we haven’t seen before that we hope will persuade AI skeptics to give this a real shot.
As Roberto Nickson said (and we truly couldn’t say it better ourselves), it is a “multi-player meme creation machine”, and that could be exactly what makes it work. And if it does take off, as Sam Altman says, we could see a “cambrian explosion” for creativity.
