AI-generated video incorporates the entire complexity of image generation, while also adding the challenge of maintaining visual coherence and physical realism throughout character, object, or camera movements. This is a challenge that OpenAI has been able to tackle with remarkable success through Sora 2, its new AI-based video and audio generation model.
In addition to introducing Sora 2, OpenAI has also announced the launch of the Sora app, an application that combines AI video creation with a social space where users can connect with others and share the content they create. For now, Sora app is available only for iOS in the United States and Canada, with access granted by invitation.
Additionally, to access Sora 2 via the web, users must have previously obtained access to the Sora app. For Android users, they will be able to use Sora 2 through sora.com if someone who already has access provides them with an invitation code.
Sora 2 is here. pic.twitter.com/hy95wDM5nB — OpenAI (@OpenAI) September 30, 2025
Sora 2 is here. pic.twitter.com/hy95wDM5nB
— OpenAI (@OpenAI) September 30, 2025
OpenAI announced Sora, its first generative AI video tool, in February 2024. Now, the technology company has unveiled its evolution: Sora 2. This new model is more precise, more realistic, and easier to control, and includes the ability to generate synchronized dialogue and sound effects to create a comprehensive audiovisual piece.
The company recognizes that it is still far from building a perfect AI video generator but highlights the fact that Sora 2 complies with the laws of physics much better than its previous models.
“Sora 2 can accomplish tasks that are exceptionally difficult, and in some cases, impossible for earlier video generation models: Olympic gymnastics routines, backflips on a paddle board that accurately model the dynamics of buoyancy and stiffness, and triple axels while a cat holds on for dear life,” explains OpenAI.
Thanks to these improvements, if you ask Sora 2 to generate a video of a basketball player missing a shot, the AI will create an audiovisual piece in which the ball bounces off the backboard or the rim in a realistic manner. By contrast, earlier models might have generated a video in which the ball spontaneously teleported to the rim, distorting objects and the actual trajectory.
Sora 2 also marks a significant leap in control. The user can enter complex instructions regarding camera movements, shot changes, etc., and the model will generate a multi-shot video that matches what was specified in the prompt.
Furthermore, the AI is able to maintain visual coherence within the generated universe as well as consistency of main characters and objects. According to OpenAI, Sora 2 excels at generating videos in realistic, cinematic, and anime styles.
Another significant advantage of this model is its capacity to generate audio and naturally integrate it within the video. Thanks to its general-purpose video and audio generation system, Sora 2 can produce background sounds, dialogue, and sound effects with a high degree of realism.
Lastly, as a new feature, Sora 2 now allows users to integrate real-world elements into the videos created using its AI. For example, you may upload your own video and ask Sora 2 to insert you into an underwater world. The tool will fulfill your request while maintaining an accurate representation of your appearance, and even your voice. This feature works with any person, animal, or object.
“The model is far from perfect and makes many mistakes, but it validates the view that further scaling up neural networks with video data will bring us closer to simulating reality,” stated OpenAI.
The Sora app integrates Sora 2 technology and offers a social experience wherein users can generate videos with AI, share them, and browse other users’ content for entertainment and inspiration. In essence, OpenAI seeks to create its own TikTok.
One of the most prominent features is “Cameo”, which enables a real person to star in a video created with Sora, maintaining high fidelity to their features and appearance. To use it, you simply need to upload a short video and audio recording so that the app can verify your identity and capture your image.
When developing this app, OpenAI has considered various issues and improper uses that could arise. For this reason, the company has implemented safeguards for disturbing or illegal content, measures against deepfakes, daily usage limits for teenagers, configuration of access and content usage settings, and periodic reviews regarding the impact on users’ well-being and mood.
The company’s objective with the Sora app is to “optimize user satisfaction in the long term,” a rather ambitious (and perhaps somewhat intrusive) goal, as Sam Altman, the company’s CEO, stated: “Most users, when reflecting on the last six months, should feel that their life is better using Sora than if they had not done so. If not, we will implement significant changes (and if we cannot resolve it, we will discontinue the service).”
What is clear is that OpenAI seems focused on creating an ecosystem that occupies as much of its users’ time as possible. On one hand, the company wants ChatGPT to become the space where users access information (even proactively delivering it to their inbox), create graphic material with its AI image generator, and carry out their purchases. On the other hand, the company aims for the Sora app to become their primary source of audiovisual entertainment.
Photo: Sora 2
Your email address will not be published. Required fields are marked *
Δ