I don’t know what sparks the initial idea for a painting in other artists. For me, oftentimes, it has nothing to do with art. As someone genetically at high risk for diabetes, the only way I could justify buying a bag of cookies was to tell myself, “I’m going to use them in a painting!” 

And so it began. Adding a few related items – a cookie cutter, a mug, a wooden table -I threw the ingredients into the AI pot of MidJourney. Among the results it generated, one caught my eyes. 

Using it as a guide, I set up my own reference: a small plate to hold the cookies, the new mug I just acquired from a craft show, and a potted plant I picked up from Home Depot. However, I didn’t care much about the background I devised. Why didn’t I just borrow MidJourney’s! I liked the idea of a painting hanging behind the objects, but I didn’t want it to feel generic. One of my cookie cutters was cat-shaped, so to add some fun, I featured a wooden mouse in the painting. The mouse is my zodiac sign, and the little wood carving was a gift from my daughter. This is how a still life became a self-portrait! 

When mixing reality with “fantasy,” lighting is the tricky part. I placed a light source on the right, but whether it replicated the effect in the AI-generated image is a question mark. Whether the lighting in the AI generation was accurate to begin with is an even bigger question mark. I decided to make the painting less about light and shadow!

After the plant’s leaves grew bigger and shifted positions, and after the cookies were replaced several times, I finally completed the painting.

The cookies were actually durable enough, but how else could I nibble an entire bag away without guilt? Though the painting is not strictly realistic, its atmosphere and staging accurately reflects my mood during the process. The objects were dear to my heart and the whimsical dynamism is quintessentially me. I’m grateful to live in a time with more tools to find inspiration and support in creating art.

PS: 

MidJourney has come a long way since I first used it, and I recently ran another comparison test by revisiting some old prompts. (Please see my first and second tests. )

For “oil painting, still life, bronze vase, light pink roses, curtain, table, realism, expressive strokes, zorn palette,” now I got these:

For “kandinsky with expressive bold strokes, fish, abstract colors:”

For “André Masson drawing, colored pencil, street musicians, metro, gloomy:”

This isn’t entirely a fair comparison because, as the model becomes more sophisticated, there are more ways to manipulate prompts for varied results. If you are willing to spend some time rating images, MidJourney builds a profile of your preferences, so the results start reflecting your taste, to some extent, regardless of the prompt.

You can also add style references to prompt for more control over the generated style:

You can even edit the result to your liking – not quite Photoshop yet, but the result can be wild. 

What’s interesting is that, when comparing the Kandinsky and Masson results, it’s not always clear that the newer models are better.