This is pretty incredible.
We've been used to waiting for Sora to arrive from OpenAI - basically we were dazzled by their potential back in February last year and have been salivating over the prospect of getting our hands on it ever since. However, it took so long - until last month - that others have overtaken it in detail, quality and complexity.
Take Google's Veo 2, for example. Now, granted, this isn't out yet - and perhaps it, too, will take ten months to arrive - but it is incredibly good. One of the reasons that it's able to do things that Sora can't is that it can be very clear that it's using YouTube videos to train on. Google's parent company Alphabet, of course, owns YouTube too - so there's no chance of a brutal court case being directed its way from the world's most used video site.
However, OpenAI's Sora has to be a little more circumspect - famously, their previous CTO Mira Morati feigned not to know whether it had been trained on any videos from YouTube. But it's almost impossible to believe that it CAN'T have used YouTube as well - there just aren't enough videos elsewhere to make them any good - but it will take a legal battle to find out for sure and it's not one that Google seems keen to undertake just yet.
Meantime, we can wait for what is certain to be a year of one-upping each other across the Text-To-Video (TTV) world. Whoever wins out as a platform, we're all going to get a lot better at prompting for great motion content - and video production costs are going to go right down.
Comments