Ideation, Inspiration, and Execution: playing with AI Art

My artist friends recently started using Midjourney to inspire their creations. And the art looks amazing.

What could a non-artist (aka me) do with AI art? I wondered, and gave it a try to create an image of a “polar bear wearing a party hat on a bicycle with watermelon wheels in Antarctica”. I was picturing something like this:

Using components available online, I created this vision sketch

I can instruct Midjourney AI on Discord just like chatting with a friend: just type “/imagine” and Midjourney will do the magic. I learned that creating AI art starts with learning to speak in ways that AI would comprehend.

I didn’t realize that my initial prompt contains a typo (it should be “bicycle” not “bycyle”), but Midjourney got it regardless. The riding motion looked odd, and there seemed to be confusion between watermelon wheels and party hat.

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicyle_with_w_b23dd50d-2222-4f42-8ac7-1d0368b3571c.png

bubblybubbles_polar_bear_wearing_a_party_hat_riding_a_bicycle_w_f0b273ba-331f-4bc5-a800-8f1d22dbbb40.png

My second attempt removed the modifier of “watermelon”. AI seemed to get the bear and the party, but still struggled with “riding”.

“*Polar bear wearing a party hat riding a bicycle”, by Midjourney*

Does AI understand “riding”? In my next attempt, I further simplified the prompt to “polar bear riding a bicycle”. One of the four initial renderings showed both bear and bicycle, which was an improvement, so I required Midjourney to create more variations. As it turned out, “riding” seemed to be beyond AI’s comprehension.

bubblybubbles_polar_bear_riding_a_bicycle_8c99ea58-b122-4bd4-b034-1b42fb70ac24.png

bubblybubbles_polar_bear_riding_a_bicycle_8cb25042-2a7a-4e6a-82bb-1379a6b2f965.png

Let’s try it differently. How about “polar bear ‘on’ a bicycle”? Would AI get it this time?

bubblybubbles_polar_bear_on_a_bicycle_40a77690-fc80-409e-89f9-240083ad0524.png

bubblybubbles_polar_bear_on_a_bicycle_1c841b84-c662-4580-92de-28eacccf11e9.png

It seemed to have gotten it! So now let’s add back the party hat. Midjourney got creative. Maybe it interpreted the prompt as “polar bear wearing a ‘party hat on a bicycle’”?

I rendered a few variations of #3, and this time it’s quite delightful!

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicycle_8a68ce5a-c049-4366-915f-37bf316c6cb6.png

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicycle_3194a7fb-4d4b-4fd4-a956-fdc1985fcfdf.png

Now, let’s bring the watermelon back!

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicycle_wheel_dc953112-b48b-4310-aa3d-aefb9c4bbb9d.png

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicycle_wheel_fc1d8e40-7b74-4f1c-b15f-b1002d7b3973.png

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicycle_wheel_3bb6475f-f1ff-4d2b-a975-12ae5b2db13f.png

bubblybubbles_polar_bear_wearing_a_party_hat_on_a_bicycle_wheel_581fd89f-0add-4d6c-9853-2ccb773dbeac.png

I didn’t get Midjourney to create exactly what I envisioned of a polar bear riding a watermelon bicycle. In another words, what is obvious to a 5 year old might not be straightforward to AI, yet.

But the course of exploration was full of delightful surprises. Together with AI, I created something that went beyond my imagination, and a few pieces of art I’d print out and frame on my wall.