OpenAI releases new Sora demo videos — with a few less glitches
OpenAI releases new Sora demo videos — with a few less glitches
Sora is OpenAI’s text-to-video generator. Two more one-minute demo reels came out last week, created by professional producers. [Tom’s Guide; YouTube; YouTube] Introduced in February, Sora is a dif…
rather “ClosedAI managed to steal even more data to train their Model to be slightly better”
The youtube thumbnail landing squarely in the middle of the uncanny valley wasn't a great start. And then I pressed played. Why did I press play?
Billions of dollars for this lol.
The videos are a complete worked example of the difference between an impressive demo - and they are impressive! - and a production tool.
Yeah you may have a point, if this came out in like 2010 as like a one-off music video instead of a venture capitalist fever dream everyone would be calling it pretty wild. Any artistic merit came from the training data, prompting, or editing together reams of raw output, rather than the shots themselves; and we all know AI companies like to dress up their demos to make them more impressive on the surface.
I think I've just gotten sick and tired of the (lack of) style by this point. It feels like staring into digital noise dressed up in a human skin. Something about the uncanny valley aspects makes it extremely uncomfortable for me to watch; whenever I watch AI videos my brain keeps telling me to run from the demons in the picture and that something is terribly wrong.
I hear 90% of the dev time was spent adding "please don't generate any elsagate content" to the prompt over and over
Who cares? It's garbage