Sora

Transform your text prompts into immersive videos with Sora’s AI-driven technology

Sora, developed by OpenAI, is an advanced AI platform that enables users to generate photorealistic videos from simple text prompts. By leveraging state-of-the-art machine learning models, Sora allows creators to animate stories, visualize ideas, and bring concepts to life with unprecedented ease.

Whether or not Sora’s hotly anticipated release lived up to the hype is up for debate. Preview clips had been floating around for months by the time it was released and excited talk of it being a truly game changing tool gave it a lot to live up to. In the event, Sora’s reception was mixed – sure it was impressive but many of the same flaws that have dogged early AI video generation tech were still in evidence.

Flashes of stunning verisimilitude are bogged down by physically unconvincing approximations of movement – AI still struggles to ‘get’ the nuts and bolts of physics. Object permanence – a cognitive concept referring to the understanding that objects continue to exist even when they are not visible – continues to be a readily discernible issue. AI video generators like Sora process frames or scenes based on individual inputs or localized data, which limits their ability to maintain a consistent understanding of object positions across time.

Maintaining object permanence requires an understanding of real-world physics, such as trajectories, occlusions, and relative movements, which current models simulate but do not inherently ‘understand’.

It seems certain that such challenges will be at least partially overcome sooner rather than later and early examples of Google’s new text-to-video generator, Veo 2, offer evidence of improving spatiotemporal models. Indeed, for the time being, it’s hard not to conclude that Veo 2’s arrival puts Sora in the shade.