Existing Retrieval Augmented Generation solutions struggle when they deal with narrative texts. This kind of texts comes with additional levels of complexity, starting from non-linear narration, through diversity of characters, to hidden meanings carried by form and language. This paper proposes approach to overcome these obstacles by incorporating text aspects described in theory of literature into current state-of-the-art solutions. The improvement is achieved by utilizing meticulously designed graph-based knowledge layout, where each important aspect of narrative text is appropriately reflected. The layout consists of multiple layers, which empowers ability to capture inner meanings of the text on different levels of abstraction and in different aspects. This article proposes complete technical solution that enables efficient work with narrative texts. Additionally, it provides a study on the most effective configuration of the framework and the importance of each aspect in the source texts. Experiments have shown that for narrative texts our approach significantly improves current state-of-the-art results.