OpenAI has released Sora 2, the next generation of its video + audio generation system, aiming to elevate realism, control, and expressiveness. The original Sora (from Feb 2024) proved that basic video generation could work; Sora 2 pushes boundaries in simulating physics, world continuity, and narrative fidelity.

Highlights & Capabilities:

• More physically realistic outcomes, e.g. bouncing balls, proper momentum, object permanence.
• Enhanced controllability: it can follow multi-shot instructions and maintain consistent world state.
• Supports mixing real and generative content: you can inject yourself or objects into scenes with accurate appearance & voice via “cameos.”
• Generates synchronized dialogue, sound effects, and background audio, not just visuals.

To showcase Sora 2, OpenAI launched a new Sora app for iOS:

• You can create and remix videos.
• The app includes invite-based rollout, parental controls, and consent for cameos (you control who can use your likeness).
• The app’s feed emphasizes creation over consumption, offers safety tools, and isn’t optimized for “time spent.”

Sora 2 is launching for free initially (with usage limits), with plans to expand access via web, API, and a “Pro” version for higher quality output.

SOURCE: https://openai.com/index/sora-2/