Generative text-to-image models, such as Stable Diffusion, allow users to easily generate images based on a textual description, the prompt. Recent developments around such models have sparked many applications and promise accessibility to a wide audience. This talk gives an introduction to the technical concept of diffusion models, which slowly generate the desired result from an initial noise pattern, taking into account the textual prompt. The practice of prompt engineering is addressed through the introduction of the Infinite Index. Based on recent research, user-centered methods for manipulating the generated output are outlined. This gives the user control over the generated output beyond prompt engineering. Implications for feed-based platforms are discussed in the context of possible application scenarios.

Niklas Deckers is a research assistant and PhD student at the Intelligent Language Technologies group (Prof. Martin Potthast) at Leipzig University and at ScaDS.AI. His research focuses on multimodal models for text and image data, including work with large web datasets.

Der Vortrag ist auf englisch. Er ist Teil der interdisziplinären Vortragreihe Feeds & Flows – Interdisciplinary Perspectives on Ephemeral Image Cultures

Montag 19 Uhr, UB Lounge