OpenAI's new text-to-video AI model Sora can create some very realistic scenes. How does this generative AI tech work?