In recent years, digital computational methods have become important tools for the study of visual arts, offering new ways to analyze and interpret artworks. However, these approaches often lack the depth and critical sensitivity provided by human perspectives. Even today, there is a shortage of computational models that balance technical robustness with critical engagement rooted in the expertise of historians, art theorists, and artists. This seminar explores the possibilities of generative artificial intelligence and advanced language models in the production and analysis of synthetic images based on museum collections. AI-generated images can serve as valuable resources for investigating patterns, gaps, and exclusions in museum archives. At the same time, using images of artworks to train generative AI opens new perspectives for the development of computer vision models that are not only more sensitive to artistic contexts but also more experimental. Thus, museums will be understood here as Latent Museums, in reference to the term Latent Space in generative AI; a compressed, abstract representation of data in a machine learning model, where similar features are grouped, enabling tasks like interpolation, generation, and clustering– but also many artificial hallucinations. Students will work with museum collections from Germany, Finland, and Brazil, providing a diverse framework to critically and creatively rethink the processes of image synthesis and manipulation in the age of AI.