Unveiling Hidden Art with AI
Generating art with AI from a painting by John Baldessari.
A friend of mine visited an art musem in Basel, Switzerland and found an interesting painting from the American conceptual artist John Baldessari.
The painting is from 1968, and describes a scene without any picture, leaving the viewer to imagine it.
The painting itself is just a blank canvas with the following text:
SEMI-CLOSE-UP OF GIRL BY GERANIUM ( SOFT VIEW) FINISHES WATERING IT - EXAMINES PLANT TO SEE IF IT HAS ANY SIGNS OF GROWTH, FINDS SLIGHT EVIDENCE - SMILES - ONE PART IS SAGGING - SHE RUNS FINGERS ALONG IT - RAISES HAND OVER PLANT TO ENCOURAGE IT TO GROW.
The words in the painting reminded him of an AI prompt, so he asked me to generate the a painting from the text, given my current interest/obsession with AI.
I copied the text from the picture, cleaned it up and gave it to Midjourney Discord as prompt.
Midjourney created 4 different variations of the painting:
The results looked great but we couldn't decide which one was the best, so we generated some variations of the first and second results.
We finally found the version we liked the most (bottom left), so I asked for an Upscale to Midjourney and we got this.
My friend was very happy with the result.
We finally know what was on John Baldessari's mind when he painted those terribly dry words.
As a bonus, I also tried to generate a painting using OpenAI's Dall-E. The results were not as artsy and good as Midjourney, but much more realistic.
Different models have different strengths and weaknesses, and Midjourney is the best tool for generating art.
There are similarities as well, both models chose to portait the girl as a very young girl not a teenager or an adult.