Makarem, SarahRöddiger, TobiasRiedel, TillBeigl, Michael2024-08-212024-08-212024https://dl.gi.de/handle/20.500.12116/44404In today’s data-driven world, effective data visualization is crucial for communication. Recent studies have shown that meaningful and relevant visual embellishments and decorations can significantly enhance data visualization memorability and comprehension. Hence, we introduce PictographAI, a generative tool integrated into presentation software to transform traditional bar charts into Pictographic visualizations. Utilizing a multimodal AI pipeline, PictographAI processes text, images, and raw data from presentation slides to automatically generate contextually appropriate pictographs. Our pipeline uses an Large language model agent, a text-guided image-inpainting model, and algorithmic post-processing to make sense of the slide contents and generate pictographs. As users update their presentation slides, the AI pipeline automates the generation of new pictographs that represent the respective contents. In this work, we demonstrate the concept and working principle that motivates the system architecture and the generative AI pipeline on a bar chart generation use case that integrates into a presentation slide creation workflow.enhttp://purl.org/eprint/accessRights/RestrictedAccesspictographsgenerative AILLM agentinpaintingvisulizationimage generationpresentationsText/Conference Paper10.18420/muc2024-mci-demo-305