I was standing in a restaurant, waiting to pay my bill when early Spring sunlight broke through cloud and lit the room. The scene was evocative of the work of the Flemish or Dutch masters - Pieter de Hooch or Jan Vemeer, maybe with a little Vilhelm Hammershøi (I know he's Danish). My curiosity piqued, I thought I'd see how a GenAI interpretation of the scene would render.
I didn't pursue this exercise with a huge degree of academic endeavour, I just used GPT 4.0 for convenience and a single prompt strategy. Capturing the scene using my iPhone - I made a quick exposure adjustment to ensure the highlights were not blown out given the intensity of the light on the subject. I then made contrast adjustments to establish a suitable reference image for my prompt and to steer the LLM in the right direction.
I could have continued to refine the image and would have dodged and burned the other patrons to give a sense of their presence without explicitly rendering them - the point of increasing contrast in my reference image. Given the environmental impact of LLM usage, I reasoned it wasn't worthwhile for the purposes of an experiment and curiosity. Besides, there is enough AI slop on the Internet already.
I wonder what Vermeer would have made of Gen AI given he was a known user of the camera obscura.
