Sora turns text prompts into video by treating visuals like a sequence of video patches. It uses a type of transformer architecture – the same family of tech that powers ChatGPT – but now it learns ...