为什么一个提示词不能生成电影：Sora的局限

In generative AI, the curse of any prompt lies within its words, forming the walls. Elaborating a bit: Recalling some Sora videos, the prompts used for generation always confined the videos within the framework of the individual prompt words. Each frame was confined to those prompt words, not deviating from them, which limited story creation. This means a 50-second video is no different from a 5-second version of the same. Unless a fundamentally different technique emerges, creating a short movie from a single prompt is not yet possible. The workaround is to divide the story into a set of prompts and then stitch them together, which is inefficient. This also applies to LLMs to an extent but can be easily overcome with better prompting. For images, the concept of weights works well. (As in Midjourney)