Sora 2 is here
OpenAI ● Covered by 3 sources
OpenAI just dropped Sora 2, a new video model that adds voices and sound effects to its clips. It's also more physically consistent, meaning fewer melting faces and warped hands.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has pushed out Sora 2, the follow-up to its original text-to-video system, and the headline upgrade is audio. Earlier Sora clips were silent, oddly dreamlike loops. This version generates synchronized dialogue and sound effects alongside the visuals, so a generated scene can now actually sound like the thing it's showing.
The company says the model is also more physically accurate than its predecessor. That's a polite way of describing the long-standing problem with AI video: objects sliding through walls, limbs bending the wrong way, water that doesn't behave like water. OpenAI claims Sora 2 handles those physics-heavy details with more consistency, which, if true, closes one of the more obvious gaps between AI-generated footage and something a human might actually shoot.
Controllability is the third piece OpenAI is emphasizing. Rather than just typing a prompt and hoping for the best, users get more say over how a scene unfolds — camera behavior, action, timing. That's the difference between a novelty generator and something creators might actually build workflows around.
All of this ships inside a new standalone Sora app, not bolted onto ChatGPT. OpenAI is clearly betting that video generation deserves its own dedicated space, separate from text and image tools, which says something about how seriously it's treating this as a product category rather than a demo.
My take — AI-written commentary, not fact-checked reporting
Audio-synced AI video is the part that'll actually freak people out, not the physics fixes — sound is what makes fake footage feel real. I'd bet this lands in disinformation debates within weeks, and OpenAI shipping it as a slick standalone app instead of a research preview tells you they've already made peace with that trade-off.
Read more about this at: OpenAI