I want to see what Stable Diffusion does with 50 years of my paintings, dataset now at 5,400 downloads

New York-based figurative painter Michael Hafftka, whose work resides in institutions like the Met and MoMA, has taken an unprecedented step by open-sourcing his artistic legacy. He has published his complete "catalog raisonné"—a comprehensive scholarly record of an artist's work—as a dataset on the AI platform Hugging Face. This dataset, spanning from the 1970s to the present, documents roughly 3,000 to 4,000 paintings and works on paper, all centered on the human figure. Released under a Creative Commons Attribution-NonCommercial 4.0 license, it is free for non-commercial use and has already been downloaded over 5,400 times, indicating significant interest from the AI research and digital art communities.

Hafftka's dataset is uniquely valuable for machine learning because it represents a "single-artist dataset" with a "consistent hand" across five decades. This provides a coherent visual language for AI models to learn from, as opposed to the sprawling, multi-artist datasets typically used for training. The artist has explicitly stated he wants to see what generative AI models, specifically Stable Diffusion, produce when trained exclusively on his body of work. He is inviting developers and artists to fine-tune models with his data and publicly share their outputs, creating a novel feedback loop between a human artist's lifetime of output and AI-generated interpretations.