Paper: HF, arXiv, GitHubContacts:
sergio.picascia@unimi.it
This research investigates how text-to-image diffusion models internally represent artistic
concepts like content and style when generating artworks. We generated 16000 images,
varying prompt template, content and style, with StableDiffusionXL. Using cross-attention
analysis, we examine how these models separate content-describing and… See the full description on the dataset page:
https://huggingface.co/datasets/sergiopicascia/thecowofrembrandt.
