OpenAI releases Sora 2 with realistic video and audio capabilities

OpenAI today announced Sora 2, the next generation of its video and audio model. Sora 2 creates more realistic and physically accurate videos than previous systems and combines them with synchronized dialogue, sound effects and background noise for the first time.

The new model also marks the launch of the Sora app, a social platform where users can create and remix content and insert themselves or friends into scenes via so-called"cameos".

The leap from Sora 1 to Sora 2

The original Sora model from February 2024 was often compared to the GPT 1 moment for video. Initial advances such as object permanence showed the potential of video AI back then. With Sora 2, OpenAI is now talking about a GPT 3.5 moment:

• Complex movement sequences such as gymnastics exercises, jumps on water surfaces or figure skating are simulated in a physically correct way.
• Errors are displayed realistically, such as bouncing balls instead of teleporting objects.
• Sora 2 is stylistically versatile: from realistic to cinematic to anime.

More control and realistic soundscapes

Sora 2 offers significantly more controllability. The model can implement multi-part scenes with consistent world logic. A new feature is the integration of audio, including voices, dialog, sound effects and entire soundscapes.

Sora 2 also allows elements of the real world to be integrated, for example by authentically projecting people into new environments using short videos, including their voice and appearance.

The new Sora app

OpenAI is launching the Sora app for iOS alongside the model. Users can discover videos, generate their own scenes and share cameos with friends. The app is designed to focus on creativity rather than consumption:

• Content is based on a personalizable feed.
• Cameos can be controlled or withdrawn by the user at any time.
• Special protection mechanisms apply to young people, including limits on video views and stricter cameo approvals.

Responsibility and security

OpenAI emphasizes that it pays particular attention to user well-being and security during the rollout. According to the company, this includes

• Transparent control options for the feed,
• Surveys on the well-being of users,
• increased moderation in the event of possible cases of abuse,
• parental controls for teenagers.

In addition, the business model is not geared towards maximum dwell time: Monetization is initially only planned via optional additional generation.

Availability

The Sora app will initially be launched in the USA and Canada, with other countries to follow soon. Users can register by invitation. Sora 1 Turbo will remain available in parallel. For ChatGPT Pro subscribers, there will also be a Sora 2 Pro version with higher quality, both online and soon in the app.

OpenAI sees Sora 2 as an important step towards general world simulators. From the company's point of view, this is a key issue on the way to AI systems that can operate in the physical world. In the short term, however, Sora 2 is primarily intended to enable creativity, fun and social interaction.

OpenAI's move towards its own social media platform is particularly exciting. While established networks are coming under increasing criticism and are struggling with stagnating growth or user migration, Sora 2 opens up a completely new usage scenario: not only professional creators, but also every normal user can transform their ideas and fantasies into impressive moving images - quickly, creatively and with the lowest barrier to entry. This could fundamentally change the way we understand social media.