OpenAI has unveiled its latest artificial intelligence program, an astonishing new text-to-video tool that it has named Sora.
While this isn't the first time we've seen programs capable of converting text prompts into videos, the quality of the samples offered up by OpenAI so far is hugely impressive.
To underline this, and to demonstrate that the sample clips it produced were not faked, OpenAI CEO Sam Altman took to X (formerly Twitter) yesterday to ask users for prompts of their own.
He selected some examples and then posted the videos that Sora created.
Advert
The very first prompt he selected came from a user asking for "a wizard wearing a pointed hat and a blue robe with white stars casting a spell that shoots lightning from his hand and holding an old tome in his other hand".
True to expectations, the clip that Altman then replied with shows pretty much exactly that, down to every detail.
Another video was created from a prompt describing "a street-level tour through a futuristic city which in harmony with nature and also simultaneously cyberpunk / high-tech".
Advert
The city that it zooms through is exactly how that sounds, looking like concept art from a science fiction movie.
Altman produced plenty of these clips throughout the day, in response to prompts as diverse as "two golden retrievers podcasting on top of a mountain" and "a bicycle race on ocean with different animals as athletes riding the bicycles with drone camera view".
They're all incredibly impressive and underline the fact that we're rapidly zooming toward a point when AI-generated video is extremely hard to spot.
Advert
For now, as mind-blowing as some of Sora's output is, that point hasn't quite been reached. There are still pretty frequent moments of weirdness around transitions and movement that betray the AI behind the video.
Still, the fidelity is a massive leap forward compared to examples from just a few months ago.
Sora isn't available publicly yet, and OpenAI hasn't said when the tool will become available for everyone, either.
Advert
What it has said is that it is "working with red teamers — domain experts in areas like misinformation, hateful content, and bias — who will be adversarially testing the mode. We’re also building tools to help detect misleading content such as a detection classifier that can tell when a video was generated by Sora".
This is crucial in the eyes of many observers, who are concerned about the possibility of faked AI videos fooling people on a variety of topics.