How does artificial intelligence intepret different artistic styles?
I’ve been creating images with Midjourney for about two weeks now and I’ll almost always add “in the style of <artist>” to the text prompt. Sometimes, I’ll get results that are spot on like the Pixar cyber pirate that I shared the other day. Other times, it seems the AI just ignores me and does what it wants.
I realize there are probably a lot of documented, technical explanations of why Midjourney works this way, but I’m on a journey of self discovery, so bear with me. For this exercise, I asked Midjourney to create variations of the photo below in the styles of Joan Miró, Frida Kahlo, Thomas Kinkade and Takehiko Inoue, four artists with very different styles.
“The base image, with a text prompt of: “photography, a storeowner talking to a customer at the cash register while a thunderstorm builds in the window behind him, finely detailed, 3d render, hyper realism, dynamic lighting”
[For each artist, I upscaled the first variation. I then upscaled each image four times, remastered it, upscaled it again, asked for a variation with “unreal engine 5” added to the text prompt, and then upscaled the result one last time.]
I loved the results.
The Store (in the style of Takehiko Inoue)
The Store (in the style of Frida Kahlo)
The Store (in the style of Thomas Kinkade)
The Store (in the style of Joan Miró)
It’s fascinating how the works diverged, yet maintain an intrinsic connection to each other.
The Inoue version is a decent take on his Japanese manga style. The scene makes me think of a ramen counter. Likewise, the Kinkade version emulates his trademark pastels and idyllic settings.
I love the Kahlo version, which looks out on what is presumably a Mexican neighborhood. The pastries on the table are warm and welcoming. The dresses and flowers in the hair are a fun tribute.
Finally, the Miró version. I stared at this for a long time, trying to figure out what the AI did here. Is it a lunch counter? What stylistic cues am I missing that tie it to Miró? Maybe the surrealness of it all? I’m not really sure. Yet, I kind of dig it.
It’s also interesting that the figures in all the versions have their backs to us. What I’ve seen is that unless you describe in detail what the characters look like, including their facial features, the AI usually defaults to the shot from behind perspective.
For additional context, the images below show two images from each artist during the experiment – the first upscaled image and another from the middle of the process. In order they are Miró, Kahlo, Kinkade and Inoue.
I did a couple similar exercises with artists and static headshots that yielded more subtle results, with maybe two exceptions.
You can see those on my website: Part 1, Part 2.
If you have tips on how to incorporate artist styles into the prompts, I’d love to hear them.
All images created by Joe Newman with Midjourney 4.
