Faculty researcher introduces concept of ‘rephotography’ to examine how AI ‘sees’ images
How do artificial intelligence (AI) systems generate images and what does their visual logic reveal about authorship, bias and historical context?
These questions sit at the heart of a new article co-authored by Dr Jim Brogden in the School of Media and Communication and Professor Kris Erickson (University of Glasgow), which presents an experimental, practice-based investigation into generative image systems and their aesthetic and cultural assumptions.
Published in the forthcoming special issue of the Journal of Visual Art Practice, the article, ‘On the Process of Image Creation with Artificial Intelligence: Rephotography Using Midjourney AI’ introduces rephotography as a critical method for engaging with AI image generation.
Rather than producing novel imagery, Brogden and Erickson use carefully constructed prompts to reconstruct canonical photographs using the generative AI tool Midjourney. By attempting to ‘rephotograph’ existing images, the researchers are able to observe how the system interprets, distorts and reassembles visual culture.
Untitled self portrait (circa 1960s) by Vivian Maier. Credit: Vivian Maier.
Through a series of empirical experiments, the article reveals recurring patterns in AI-generated images, including:
- Hallucination
- default aesthetics
- overfitting to iconic visual forms
- a detachment from historical and social context.
Instead of treating these outcomes as technical failures, Dr Brogden and Professor Erickson frame them as productive symptoms – entry points for critique that expose how generative systems are shaped by their training data, probabilities and omissions.
Midjourney outputs at intermediate steps a-d.
“Our analysis draws attention to the layered processes involved in AI image creation, including prompt engineering, data culture and questions of authorial agency,” Dr Jim Brogden says. “In doing so, our work challenges assumptions that AI-generated images are neutral or transparent representations. Generative images are instead saturated with the biases, absences and logics embedded within their training corpora.
“Against this backdrop, rephotography functions as a performative and critical mirror, reflecting the system’s underlying visual assumptions back to the viewer.”
The work also situates contemporary AI image practices within longer histories of photography and visual culture. Drawing on discussions of canonical photographic self-portraiture – including the work of Vivian Maier – the article highlights how framing, omission and context shape meaning, and how these concerns persist, albeit in altered form, within AI-generated imagery.
