IMHO, text-to-video generation, or text-to-video conversion, is an unsolved problem; therefore, such a system does not yet exist. Take for example The Lord of the Rings (film series), I can say for sure that the movies Peter Jackson made were absolutely, categorically not the movies I saw in my head when reading those books. Thus, there will be no one right way to accomplish this, making development of any such system a Slippery slope.
I have asserted before on Quora that such a system might be made from analyzing the entirety of YouTube, for instance coupling the text of automatic captioning with frame-by-frame images; however, I do not believe even Google has the capacity for doing this yet. Beyond that, then there is the issue of reversing the process of analysis for generation. So, I suppose this is theoretically possible. I can imagine that by the time such processing infrastructure is readily available the machine learning technology of today may have changed as well.
In short, to address your rambling and hypothetical question more directly, such a system would most likely need to be trained on a large number of novels within each genre, or vertical, given today’s machine learning technology.