People tend to confuse the chaotic whirlpool of their own non-verbalized thoughts, and very deterministic pattern of book text.
The most complicated book contains just a skeleton of the imagined picture, which should be created by the user itself, based on his previous experience, picture of the world, and associations.
The text is poor, it consists of limited set of synonyms, actors, and details, short sentences, connected with commas.
AI should read a human book, detect actors and details (nouns, adjectives, etc), their actions (verbs), and build an object tree.
Then it should generate an object-based plot, based on this tree. Then turn into words (or pictures) to output.