OpenAI has launched the Sora app, a TikTok competitor powered by its new video and audio generation model, Sora 2. This model is more physically accurate, realistic, and controllable than previous versions, featuring synchronized dialogue and sound effects. Sora 2 represents a significant advancement in video generation, capable of complex and physically plausible actions like Olympic gymnastics and realistic object interactions, overcoming limitations of prior models that often distorted reality.
Sora 2 Capabilities
Sora 2 models the physical world more accurately, including failures and realistic physics, such as a basketball rebounding off a backboard rather than teleporting. It can follow detailed instructions across multiple shots and excels in various visual styles, including realistic, cinematic, and anime. The model also generates sophisticated background soundscapes, speech, and sound effects, enhancing realism. Users can inject real-world elements into generated videos by uploading a short video and audio clip, allowing the model to recreate their likeness and voice in any scene.
Deployment and Social Features
The Sora app, available on iOS, enables users to create, remix, and discover videos in a customizable feed. A unique feature called "cameos" lets users insert themselves or friends into videos with high fidelity after a one-time verification recording. The app was initially launched internally at OpenAI, where it fostered new social connections. The social experience centers on creation rather than consumption, emphasizing community and interaction through invite-based access.
Responsible Launch and Safety Measures
OpenAI addresses concerns about addiction, doomscrolling, and isolation by giving users control over their feed using natural language-instructed recommender algorithms. The app prioritizes content from people users follow and videos likely to inspire creation, not prolonged viewing. Teen users have default limits on daily video generations and stricter cameo permissions. Human moderators and automated safety systems are in place to handle bullying and harmful content. Parental controls via ChatGPT allow parents to manage feed limits, personalization, and messaging.
Users maintain full control over their likeness in cameos, with the ability to revoke access or remove videos at any time. OpenAI has tackled safety issues including consent, provenance, and harmful content prevention. The app’s monetization plan is minimal and transparent, focusing on user wellbeing rather than maximizing engagement or ad revenue.
Availability and Future Plans
The Sora iOS app is rolling out in the U.S. and Canada, with plans to expand globally. It is free initially, with generous usage limits, though constrained by compute resources. ChatGPT Pro users can access a higher-quality Sora 2 Pro model. Sora 2 will also be available via API, and the original Sora 1 Turbo model remains accessible with all user content preserved.
Broader Impact
Sora 2 marks a major step toward general-purpose world simulators and AI systems capable of functioning in the physical world. OpenAI envisions these advances will accelerate human progress and foster new forms of creativity, connection, and entertainment, aiming to create a healthier platform compared to existing social media options.
