OpenAI unveils Sora 2: Next-gen AI video generator with Physics and a New Social App

OpenAI officially debuted its newest text-to-video model, Sora 2, on September 30, 2025, calling the breakthrough the "GPT-3.5 moment for video." The highly anticipated release not only delivers a significant leap in realism and control but is also launching alongside a new, short-form social media app called Sora, setting the stage for a dramatic shift in digital content creation and consumption.

Sora 2 is designed to go beyond the capabilities of previous AI video systems by focusing on advanced "world simulation."

Key upgrades: Physics and sound

The most notable improvements in Sora 2 address the core weaknesses of earlier generative video models: inconsistent physics and the lack of audio.

Accurate real-world simulation: Sora 2 demonstrates a significantly better understanding of the physical world. For example, where older models might "morph objects and deform reality" to execute a prompt (like a basketball magically teleporting to the hoop), Sora 2 renders scenes with accurate dynamics. The model can now simulate complex interactions like the buoyancy of a paddleboard or the trajectory of a missed basketball shot.

• Synchronized audio and dialogue: A major new feature is the addition of synchronized audio. Sora 2 can generate realistic background sound effects and dialogue in multiple languages that are timed correctly with the visual action.

• Controllability: The model supports far more detailed text prompts, allowing users to customize cinematic elements like the camera equipment, lens effects, and multi-shot consistency.

A new social app with "Cameos"

To deliver Sora 2 to consumers, OpenAI has launched a dedicated, invite-only Sora iOS app, a direct move into the social media space currently dominated by platforms like TikTok and Instagram Reels.

The app's standout feature is "Cameos," which allows users to insert their own verified, realistic digital likeness into any AI-generated scene. After a short, one-time video and audio recording to verify identity, users can use their Cameo to:

• Place themselves as the main character in any generated video

• Have their AI avatar deliver a scripted narration or dialogue

• Control who can use their likeness, with the ability to revoke access and review any content featuring their Cameo

The new Sora app features an algorithmic feed, customizable through natural language prompts, and includes strong parental controls accessible via ChatGPT to manage personalization and direct messaging for teens.

Availability and safety concerns

Sora 2 is initially rolling out on an invite-only basis in the U.S. and Canada via the new iOS app. OpenAI has stated the model will be free to use with "generous limits" during the initial phase, with a premium "Sora 2 Pro" model available for ChatGPT Pro subscribers. API access is also planned for the near future.

Despite the excitement, the launch has reignited debates over AI-generated content:

• Deepfake risk: The Cameo feature, while consent-based, pushes the boundaries of accessible deepfake technology

• Copyright: The model's training data has raised copyright concerns, with reports indicating content owners must actively opt out of having their work replicated in the Sora 2 video feed

• Content Moderation: In the days following the launch, some early users reportedly managed to generate videos with violence and copyrighted characters, challenging OpenAI's announced safety guardrails.

OpenAI acknowledges the risks but positions Sora 2 as a tool to "democratize creativity," stating that all generated videos will contain visible watermarks and C2PA metadata to track their AI provenance.