OpenAI releases AI text-to-video generator Sora, ponders Google Search rival
The photo at the top of this article is not real. In fact, it’s not a photo at all but a screenshot of a video generated by OpenAI’s new text-to-video model, Sora.
On Thursday, the company behind ChatGPT introduced Sora, which is an Al model capable of generating videos up to one minute long with relatively simple text prompts. OpenAI claims that Sora is able to generate “complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background.”
Most impressive of all, Sora understands how objects and environments should interact in the real world. For example, here is a clip of a woman walking down a city street in which the lights of the buildings reflect off of the water, her clothes all sway as she movies, and dozens of other people go about their business in the background:
All of this is from a single text prompt. Yes, there are some obvious signs that this isn’t real life if you look closely enough, but if you were scrolling through social media and caught a glimpse of this video, how long would it take you to realize it was AI?
For now, Sora is only available to the OpenAI Red Teaming Network, which consists of individual experts, research institutions, and civil society organizations. But it’s only a matter of time before the public gets their hands on this technology as well.
As such, OpenAI is prioritizing safety when it comes to the eventual rollout of Sora. So-called red teamers are “adversarially testing” Sora to find any serious flaws. OpenAI is also simultaneously building useful tools that will help detect misleading content generated by Sora. If and when the technology behind Sora finds its way into an OpenAI product, the company will include C2PA metadata that will make it easier to uncover its origin as well.