OpenAI Reveals Sora 2 With TikTok-Style Social App That Puts You in the Videos

Openai released Sora 2 on Tuesday, combining it with the latest video generation model with a new social app that allows users to create, share and star with clips generated by AI. The company took a major step in simulating this release physical reality, with models now generating synchronous audio along with video for the first time.

The updated model can generate video clips showing the complex physical interactions that previous systems struggled with. In some instances, Sora generated a character who refluxed into the Olympic gymnastics routine, a paddleboard, performing triple axels without obvious distortion or morphing. Unlike previous video generators that bend physics to meet text prompts, SORA 2 attempts to model realistic outcomes, including obstacles.

“Previous video models are overkill. They change objects, transform reality and perform text prompts well,” Openai said in the announcement. SORA 2 states, “It’s better about following the laws of physics compared to previous systems.”

This model generates background soundscapes, audio, and sound effects directly from the text prompt. Until now, the only model of that feature has been Google’s VEO 3. The system processes multiple shot sequences, while maintaining continuity throughout the scene changes, is extremely complex and requires a good understanding of both the character and the environment.

Openai sells the Sora 2 as “GPT-3.5 Moments in Video”, comparing it to the language model predecessor ChatGpt. The original SORA, released in February 2024, represents what the company calls the “GPT-1 Moment.” This indicates that video generation is beginning to work at scale.

Many good models quickly left Sora in the dust, so by the time Openai decided to release the model, the Chinese alternative was able to use the same prompt to output better, more coherent videos.

For now, the only way to test your model is simply to invite it through a new iOS app called SORA. Unlike previous models that can only be accessed through websites and focus on isolated video generations, the app appears to be more refined and versatile, introducing a feature called “cameo” that allows users to insert themselves into generated scenes.

After recording a short video to validate identity and capture appearance and audio, users can view it in an environment created by SORA. This feature works for humans, animals, or objects, and users control who can use the portrait.

During the demo, Openai teams performed ads, kickflips and generated videos of themselves featured in style similar to Tiktok videos and Instagram reels in a variety of situations.

The app includes a customizable feed using Openai, described as the recommended algorithm for a new class that accepts natural language instructions. The system is the default for displaying content from people users follow or interact with, and the company said it doesn’t optimize for the time spent scrolling. The built-in mechanism provides the option for users to vote regularly about their well-being and adjust their feed settings.

For teenagers, the app includes default limits for the daily generation that appear in the feed. Parents can access controls via ChATGPT to manage scrolling restrictions, algorithm personalization, and direct message settings.

Users can maintain full control over the cameo and cancel or delete videos containing likeness at any time. The app shows users all videos featuring cameos, including drafts created by other unpublished drafts.

SORA 2 will be launched in the US and Canada through an invitation-based system, and plans to expand quickly to other countries. This service is free of charge, what Openai calls “generous restrictions,” but these are subject to the calculation of constraints. ChatGpt Pro subscribers have access to an experimental high-quality version called the Sora 2 Pro. The company plans to release the Sora 2 via API, making its previous SORA 1 turbo model available.

Openai said SORA 2 will ultimately offer users the option to pay for additional generations if demand exceeds available computing resources.

For now, if you don’t have an invitation code, iPhone, or ChatGPT Pro, the only option is to use a local video generator like a limited VEO 3 run or WAN. There are also cheap options like Kling, Seedance, Hailuo and runway, but the appeal of having a very realistic video model with social media features is a plus no one else in the industry offers.