On Tuesday, OpenAI dropped Sora 2, the newest iteration of its 2024-launched video-generation tool. Figure skaters with cats on their heads, dog astronauts gobbling tennis balls, unusually agile horseback riders standing astride multiple animals — these were some of the snippets seen in its short video presentation.

Sure, the model still struggles with dialogue — notice that the scenes feature mostly voiceover when there are words at all. And slow down the videos to individual frames and you’ll see plenty of hallucinations. But OpenAI did enough shiny things Tuesday to make users take notice. The company will release — first by invite, and then presumably commercially — a tool that will let users prompt these videos into existence and share on their social feeds. Now instead of dunk

See Full Page