In this experiment, I wanted to create a portrait of the Venus of Willendorf in the style of Chaïm Soutine’s characters. I encountered Chaïm Soutine's paintings a few years ago and was deeply impressed by the expression in each one of them.
I wondered what I would get if I briefly asked an AI model to portray the Venus of Willendorf sculpture. Which elements would it preserve, and which would it extract? How would it fill in the subtracted elements, like the face or hands, for example?
I was curious whether I would recognize any societal influences in the output and how my figure would appear to move.
I did several iterations, and the most notable thing I observed is that the model struggles to imitate Chaïm Soutine's painting style. It can attempt to replicate the look of oil painting, but not the dynamic, odd rhythm and movement characteristic of his work.
I guided the model by uploading a reference image and refining the text multiple times. Two key observations stood out:
In the first iteration, my character covered her chest with her hands (even though the Venus of Willendorf sculpture lacks hands). As the rendering progressed, and after adding a reference to Chaïm Soutine, the character no longer covered her chest but instead a fabric covered her entire body, except for the chest. Eventually, the character evolved to be holding something in her hands that resembled a baby-like figure.
From left to right - Venus of Willendorf, Chaïm Soutine - Portrait of Madeleine Castaing (c. 1929) oil on canvas, and two selected works by Midjourney + a little additional editing.
READING