Fantastic summary! These Sora demos are truly mind-blowing. I'm especially intrigued by the 'foresight of many frames' part for consistency; does it leverage some form of advanced temporal embedding or a semantic scene graph? It's like those 'AI won't do creative' takes from last year are already historcal, isn't it?
Fantastic summary! These Sora demos are truly mind-blowing. I'm especially intrigued by the 'foresight of many frames' part for consistency; does it leverage some form of advanced temporal embedding or a semantic scene graph? It's like those 'AI won't do creative' takes from last year are already historcal, isn't it?