Intelligent Visualization of Book Narratives using Natural Language Processing
Visual learning aids in the form of comic strips based on literary prose require prohibitive generation costs and complicated procedures for educators. In contrast, this paper introduces Kapow! AI, which synthesizes annotated comic strips from narrative text using entirely open source resources within a typical Google Colab notebook setting equipped with a T4 GPU. The process involves three open-weight models: a BART-large-CNN summarization model, which extracts scenes from the text, Stable Diffusion v1.5 to synthesize corresponding image panels, and a PIL annotation layer which captions each image panel with the extracted scene description. Proprietary APIs and commercial services are not needed for the process. Tested on fifteen public domain texts from three fictional genres, the proposed pipeline yielded an average CLIP alignment of 0.31 and user experience score of 4.0 out of 5. The main drawback of the model is character inconsistencies between panels due to independent image generation per panel.