OpenAI's video-generating AI service, Sora, launched in December 2024, shocked the market by creating high-quality videos from text. The market evaluated it as a game-changer that would alter the paradigm of video production. Concurrently, various services such as RunwayML, Google DeepMind Veo, Pika, and Hailuo were launched, rapidly advancing the generative AI content industry.

However, Sora's technological advancement lagged compared to other services. Meanwhile, Google unveiled Veo3, capable of generating videos longer than one minute and featuring enhanced image-to-video conversion. Runway strengthened its high-resolution video generation capabilities with its fourth-generation generative model.

OpenAI, aiming to enhance its video-generating AI competitiveness, released Sora 2 on September 30, 2025. An intriguing aspect is the provision of a smartphone app in addition to the existing online service. The Sora app, launched for Apple iOS, received a positive response, ranking first in downloads on the U.S. App Store immediately after its release. Here is an exploration of what the iOS version of Sora looks like.

Sora 2 Implementing Physical Laws and Sound

Sora 2 achieved several technological advancements that surpass the limitations of the existing model. A notable change is the sophistication of the physics engine. Sora 2 more accurately understands and reflects real-world physical laws such as gravity and friction in videos. For instance, inputting a scene where “a basketball hits the rim and bounces off” results in natural movements that seem to calculate the ball's elasticity, rotation, and the angle of impact with the rim. According to OpenAI, awkward object movements or unrealistic interactions previously seen in the initial Sora model have been significantly improved.

The ability to reproduce complex movements has also been enhanced. It naturally generates gymnastics moves, backflips on a paddleboard, and triple axels in figure skating. It maintains consistency in videos even with complex commands.



The addition of synchronized audio generation is one of the changes in Sora 2. While past versions of Sora created silent videos, Sora 2 automatically generates dialogue, sound effects, and background music that match the video's content and mood. When users include directives like [dialogue] and [sfx] in their prompts, it generates dialogue synchronized with the character's lip movements and adds appropriate sound effects to scenes. The range of style choices is broad, allowing for realistic live-action videos, cinematic productions, and animated styles.

The addition of these two features is expected to enhance the immersive experience of video content. Depending on future development directions, it is likely to evolve into a 'multimedia content creation' tool. OpenAI evaluated Sora 2 as the 'GPT-3.5 of the video field.' This expression reflects confidence that, just as ChatGPT brought innovation to the conversational AI market, Sora 2 will lead groundbreaking changes in the video generation field.

The iOS App Resembles a Social Platform More Than a Video Production Tool

The Sora app for Apple iOS includes a social media feed-style interface and video generation features such as Cameo and Remix, allowing ordinary users without professional video editing skills to easily create high-quality videos.

First, the Cameo feature reflects the user's face or a specific person's appearance in the video. By scanning and registering one's face, users can create movie trailers or animations where they are the protagonist simply by entering text. OpenAI mentioned that a consent-based safety mechanism has been established to protect user portrait rights. When selecting another user for Cameo use, a message is displayed indicating that a draft will be shared.



The Sora app gives the impression of a social platform similar to short-form video feeds like TikTok and Instagram Reels. Videos created by creators running the Sora app are displayed, and other videos can be played by scrolling up and down. Videos generated by the user are shared on the feed. Additionally, OpenAI recommends posts by understanding user activities, location, post engagement, and ChatGPT conversation records. This is akin to a video algorithm generation structure, which might lead to the repeated playback of specific videos. This can be disabled by turning off 'Use personalized settings for exploration' in the app.



The Remix and collaboration features allow users to recreate videos made by others or trending videos within the app in their own style. It also supports a collaboration feature where multiple users can work on a video project together, expanding creative potential.

For testing, a video of a person walking on water was generated. By organizing and inputting settings for animation-style generation and movement in text, the process shifted to video generation. It took about one minute to generate the video. Once video generation is complete, it asks whether to post it on the feed.



Due to the limitations of smart device apps, the video quality is not clear. It is close to the 480p video quality used by free users. The video length is also short. Sora 2 can only generate short videos of up to about 20 seconds. While useful for creating short clips or social media content, it is lacking for producing longer videos with narratives.

IT Donga Reporter Kang Hyung-seok (redbk@itdonga.com)

Popular News