“Reticent Reproductions” utilizes Open AI’s ChatGPT 4o mini general purpose Large Language Model (LLM) to describe works of 2D visual art. These descriptions are then given to Adobe’s Firefly Image Model in Photoshop, using the Generative Fill Function to create images from GPT 4o’s text. The combination of these two models “recreates” the artworks described.
—-
ChatGPT 4o mini and Adobe Firefly team up to recreate visual artworks, from ubiquitous to obscure
The images produced ranged from fairly close to the original to full-blown Bullshit, a term I will be using throughout this project. Where a “hallucination” requires some kind of understanding of the truth, bullshit is a construction that is present as the truth without regard for the consequences. The works below showcase the prompt used in ChatGPT, the response from GPT, and the Image generated by Firefly side-by-side with an image of the original artwork. I have also included personal notes and thoughts about each work, and its generated counterparts. The ways in which these two systems recreate artworks illustrates the limitations of the use cases of AI to be a consumer and creator of art, as well as the consequences of automating many, or any, of the steps of the creative process.
By

Maxine Vollertsen