“Painting of a black cat with white markings sitting on a chair” is the best thing Franz Kline has painted in 60 years! Love the coiled energy and the furry tail on the left.
This is the best of 9 images that the open-source AI DALL·E Mini generated in about 150 seconds when I prompted it with a friend’s suggestion “Painting of a black cat with white markings sitting on a chair by Franz Kline”. Franz Kline died in 1962, but this looks exactly like one of his better paintings. DALL·E Mini (now renamed Craiyon) is a cheap and rough reimplementation of OpenAI’s huge DALL-E; Mini has only been trained on about 30 million images, but that’s more than enough for it to learn the style of many great artists.
Interestingly, DALL·E Mini has trouble conceiving of a Mark Rothko painting of a cat, so it tends to just add one in front:
If you train an AI on captioned images, it learns to create anything you can describe. OpenAI’s DALL-E 2 is phenomenal, the more detail you give it the more it produces. @JaredZimmerman prompted it with “lonely 1930s gas station situated by itself in the distance on a desert highway in late evening with no cars or customers and sharp shadows cast by the parking lot lights with intense contrasting underpainting in minimalist painting style with visible brush strokes,” and…. wow.
Or you can just prompt it with the name of a painer, and it will deliver. It’s been educational following the #dalle2 hashtag on Twitter to learn about French painter Florine Stettheimer, the Kalighat style of Indian painting, etc.
The objections are threadbare
Here are some of the “but it’s not art” objections that people have raised. Knocking these down is easier than falling off a log.
Art is more than photorealistic images. Definitely, but AIs generate near pixel-perfect renderings of crayon drawings, thick impasto, woodcuts, oil painting, even Boris Johnson made of play-doh. @JaredZimmerman has coaxed impressive images out of full-blown DALL-E 2, e.g. “Ukiyo-e print on heavily textured paper with strong black outlines and limited muted colors of dramatic california coastal cypress trees…” If you’re seeing art on a computer screen, you can’t tell whether you’re seeing a photo of the medium or the AI’s generated pixels resembling the medium.
Art is an expression of humanity. That definition is very species-centric and reductive, since birds construct artistic nests, there’s an elephant that makes art with paintbrushes, etc.; and even if one accepts it, it doesn’t explain why an AI that’s ingested hundreds of millions of images and captions can’t learn how humanity expresses itself in art better than any puny human?
Art must come from a full consciousness. Why? That’s close to begging the question (finally a correct use of the term!). Also, it seems there’s little connection between “full consciousness”, whatever that means, and artistic merit, considering all the mentally disabled people who make expressive art; and even if one accepts the connection, why isn’t the connection established by the person who writes the prompt “Dystopian oil painting of a cyborg repairing her firmware while hiding in a back alley in Kowloon Walled City in a rainy evening”? (resulting in one of the best DALL-E 2 images I’ve seen so far)
The artist decides what they want to say and how they want to say it, the image generator is just a tool. Fine, but now the artist is any human being, which is quite a change from what “artist” has meant for a thousand of years! My response to anyone who reacts to modern art with “I could have done that” is always Go ahead, do it! The world needs more art! Now we’re faced with limitless abundance of artistic images.
Art is something made by human hands. A digital image is great, but it’s not the same as seeing Van Gogh’s sublime thick brushstrokes up close in 3D. Folks, go to the graduate shows of your local art college and support human artists! Artists have explored and narrowed the boundary between human mark-making and mechanical production for over a century, e.g. Andy Warhol’s screen prints of Marilyn Monroe, or Robert Ryman’s reductive white paintings in which you focus on how the white surface is mounted to the art gallery’s white wall (I used to love his work, now I find it vaguely irritating), etc. However, this definition of art as “something that isn’t presented on a screen” is news to all the artists who have worked in video and computer images for 50 years.
An artist doesn’t give you what you want. He/she/they creates something that induces a feeling in you that you previously didn’t have. The main job of art is to make people think and contemplate. Nice definition, but I asked DALL·E Mini for “Painting of a black puppy by Salvador Dali” and in 150 seconds I had 9 images of evocative surrealist paintings. The best one made my friends think and contemplate, and it told a story based on my prompt. Of course it is real art!
Folks who maintain this stuff isn’t Art have to redefine “Art” to escape the uncomfortable truth that we now have incontrovertible proof that if you train a large AI model on enough images and captions, it learns how to produce genuine art that people like; “Beauty is in the eye of the beholder,” and the AIs have nailed it.
We live in interesting times!