Using GPT-4 with Vision as an Art Critic

How OpenAI’s latest model can provide insights into visual arts

minimalist painting of an AI art critic, Image created using an AI image creation program, Midjourney, and edited by the author

About two weeks ago, OpenAI started rolling out access to their GPT-4 model with vision (GPT-4V) [1]. As a subscriber to ChatGPT Plus, which costs US$20 per month, I got early access to the new model. The system is available here: chat.openai.com.

With GPT-4V, I can submit a text prompt with up to four images. The system will analyze the prompt and images and respond accordingly. Since I have been writing about the creative uses of AI and ML, my first instinct was to see if GPT-4V can effectively function as an art critic.

Here is my first experiment. I had the system review an acrylic painting of a New England landscape that I made back in 2002 with the help of a reference image, some custom software, and an overhead projector. I took a photo of my painting and sent it to GPT-4V for review.

ChatGPT User Interface (left), Warren Park in Autumn (right), Screenshot and Painting by Author

Here’s my prompt, followed by GPT-4V’s review.

RobG: Based on the photo of this painting, write a review with constructive criticism.
Warren Park in Autumn
Robert A. Gonsalves
Acrylic on Canvas Board
24” x 32”