OpenAI has introduced a new version of its video and audio generator, Sora 2. The company promises more realistic object motion physics, better control over scenes, and synchronized sound. In parallel, the Sora social app for iOS is launched with a cameo feature.
OpenAI has introduced a new version of its video and audio generator, Sora 2. The company promises more realistic object motion physics, better control over scenes, and synchronized sound. In parallel, the Sora social app for iOS is launched with a cameo feature.
In its press release, OpenAI explains that Sora 2 more accurately follows cause-and-effect relationships: a ball bounces off a backboard after a miss, a surfboard reacts to waves, and tricks more closely resemble the work of real muscles and materials. It also has noticeably better «obedience» to complex instructions across multiple frames and scenes.
The model better maintains the «state of the world» between plans: you can set a sequence of actions, change angles, but preserve objects and their properties. For creators, this means less object distortions and «teleportations» of objects during complex queries.
Sora 2 generates voice, background noise, and effects that blend with the image—from character speech to footsteps that match the movement in the frame. This reduces the need for subsequent manual voiceover.
The Sora app for iOS lets you record a short video and audio sample and add yourself or friends to any scene. The app lets you remix other people’s work, creates a personalized feed, and, according to OpenAI, is focused on creating, not «getting stuck» in viewing.
The company emphasizes feed management tools, teen restrictions, bullying moderation, control over the use of your likeness (you can revoke permission and delete videos with your cameo). Sora has systems to prevent the creation of harmful content and label the origin of materials.
The Sora app will launch in the US and Canada by invitation, with access gradually expanding. Sora 2 will initially be free with «generous limits,» but with computing resource constraints. ChatGPT Pro subscribers will be able to try an experimental version of Sora 2 Pro on sora.com (later in the app). There are also plans to release Sora 2 via API. The previous Sora 1 Turbo remains available.
In July, dev.ua wrote about how Sora 2 could be the biggest update among OpenAI’s text-to-video models.
Earlier, Microsoft Bing launched Bing Video Creator in its app, which uses OpenAI’s Sora model. Users will be able to create videos from text prompts.
