OpenAI has introduced a new artificial intelligence model called Sora, designed to generate videos based on text descriptions. While still in its development phase, Sora demonstrates significant potential in the field of video creation, with the ability to handle complex scenes and understand real-world physics. Sora can currently produce videos spanning up to a minute in length, and the model demonstrates an understanding of object interactions and can translate textual instructions into visually coherent representations. Greg Brockman of OpenAI took to X to show off a video of a woman walking in Tokyo, generated entirely by Sora.

Similarly, a video featuring a robotic arm sorting objects showcases Sora's ability to generate fluid and accurate mechanical motions. This indicates the model's potential for practical applications in fields where simulations and visual demonstrations are valuable. While promising, Sora does have some limitations. For example, it may encounter difficulties with complex physics or introduce spatial inconsistencies, and the OpenAI team is actively addressing these areas for future improvement.

OpenAI emphasizes the importance of using Sora ethically and responsibly. The company says its working with experts to proactively identify potential areas of misuse and are developing tools to detect videos generated by Sora. This approach encourages transparency and safeguards against the technology's potential for creating deceptive content, as the generation of videos is potentially rife for abuse. Sora represents a massive step forward for generative AI, as the closest to video generation previously available was the Stable Video Diffusion model, which could generate a six-second video from a photo.

The ability for Sora to produce videos can have a massive impact depending on how it's rolled out. OpenAI hasn't said how the model was trained or with what data, only assuring The New York Times that it's a combination of publicly owned videos and ones licensed from copyright holders.

Sora is not publicly available for download currently, which OpenAI tells The New York Times is because the researchers are still working to establish its dangers.