In the middle of the night, OpenAI suddenly dropped a bombshell: Sora 2 is here.
After watching the livestream, any trace of sleepiness completely vanished.
OpenAI themselves described it like this:
“With Sora 2, we are jumping straight to what we think may be the GPT-3.5 moment for video.”
Yes, the ChatGPT moment for AI video has officially arrived.But accessing Sora 2 still requires an invite code.
If you don’t have an invite code, don’t worry—I’ve tested a verified way to access Sora 2. The third-party platform Global GPT has already integrated Sora 2, and you can use it freely without any verification.
What Is Sora 2?
The all-new Sora 2 is a native video-and-audio generation model.
Notice the wording: it’s not just a video model anymore.
It’s somewhat similar to Google’s Veo 3, but based on the demos released so far, Sora 2’s overall quality goes far beyond Veo 3.
GPT-5 shook the world, and now Sora 2 feels like OpenAI’s triumphant return.
This launch actually consists of two parts:
- The Sora 2 model – which is basically the new SOTA for AI video.
- The Sora 2 App – which shows OpenAI’s even bigger ambition: to build the TikTok of the AI era. A social, creativity-driven platform.
Let’s break it down.
Part I: Sora 2 – Physics, Motion, and Realism
Right now, AI video models are all racing toward the same goals: physics, human performance, consistency, and audio.
- Physics & realism: The gold standard is sports—especially gymnastics and ball games. Most models fail here. Even state-of-the-art ones like Hailuo 02 or Keling 2.5 only manage partial success.
But Sora 2? It’s insane. It can generate Olympic-level gymnastics, backflips off paddleboards, and even volleyball matches.
👉 Prompt example of Sora2 : A gymnast flips on a balance beam, cinematic style.
Compared to Sora 1, the progress is staggering. It feels like decades passed, not just a year and a half.
compare Sora2 with the original Sora 1
And don’t forget—these aren’t just GIFs. They include sound. Volleyball smashes, splashes from dives, even ambient sound—all rendered with near-perfection.
👉 Prompt example of Sora2 : A man jumps off a diving board doing a cannonball.
From what I’ve seen, Sora 2 has set a new benchmark in motion quality and physical accuracy.
Sora 2 Human Performance & Consistency
The real battleground now is AI actors—digital humans who can speak lines with natural expression.
With Sora 2, you can verify your identity, create your own digital double, and assign it as a fixed ID for reuse.
That means you can summon your character in different videos, with consistent appearance and performance. Combined with near-perfect audio generation (voices, ambient noise, crowd chatter), AI filmmaking is no longer a fantasy—it’s within reach.
👉 Prompt example: @daniel playing the trumpet among a herd of zebras.
👉 Prompt example: @daniel and @duxin arm-wrestle—your choice who wins.
In both cases, the faces are consistent. The acting, expressions, and sound design are indistinguishable from real footage.
Sora 2 Versions: Standard and Pro
Under the hood, there are two versions: Sora 2 and Sora 2 Pro.
Right now, access is limited: priority for ChatGPT Pro users, U.S. and Canada only, and locked behind invite codes. Each early user gets four invites to share with friends.
The currently available version is, frankly, not very practical—it’s heavily restricted, offering almost no control over generation parameters.
You can only create 10-second clips in either landscape or portrait orientation, and the resolution is extremely low at just 360p.
We’ll have to wait for updates.
Part II: Sora 2 App – AI Social Platform with Cameos
This is where it gets interesting.
The web version has already launched, and the iOS version is live in the U.S. App Store (no Android yet). Access is invite-only.
At first glance, it looks like an AI-powered TikTok: scroll through public AI-generated clips, like, share, follow. Even the UI feels identical.
But the real innovation is a feature called Cameos.
Every avatar is a cameo—you can @mention them to appear in your video.
Example: I can @rockhazix (myself) and @sam, and generate a clip of us having a quirky dinner together in a stylish restaurant.
It’s essentially guest-starring your friends in your AI video. And that makes it a social app at its core, not just an AI video feed.
OpenAI emphasizes that Cameos are identity-verified. You’ll need dynamic audio checks, liveness detection, and more to prove it’s really you. Once set up, you can tweak how your cameo looks and acts.
The possibilities are wild: I generated myself and Ultraman eating dinner together. I even shouted “Sora 2 is out!” on a busy street—and the facial animation was so intense it shocked me.
That said, face consistency is better for Western features than Asian ones (a common issue with Western-trained models). And yes, it can be a little uncanny.
Sora 2 Final Thoughts – Social App Potential and Future
Let’s talk about the Sora 2 App.
Frankly, I’m skeptical. Historically, every attempt at AI video social feeds has fizzled out. Because creators want exposure—so why post on a tiny new app instead of TikTok?
And honestly, do average users even care if a video is AI or not? Most don’t. All that matters is whether the video is good.
So is an “AI TikTok” a doomed idea? Maybe.
But Sora 2 App is a little different. Thanks to Sora 2’s huge leap, OpenAI turned it into a social creativity product built on Cameos. It feels closer to Snapchat or Faceu than TikTok.
The risk is the same as all viral social apps: quick rise, quick fall. Think of BeReal in 2022—it exploded, then faded once the novelty wore off.
Still, with AI lowering the barrier to creativity, plus the fun of remixing and pranking friends, there’s potential for a vibrant ecosystem. Or it could collapse into sameness.
English