Read"> Read"> Read" />

Best of the Bay 2022

An experimental collaboration with then-new AI Image Generation technology. Can't beat it? Try and do something interesting with it.

Client: San Francisco Bay Guardian

2022

share

As a creative firm we are well-aware of the challenges that AI technology brings to our industry. In the interest of using AI as a tool rather than a competitor, we executed this experimental project. We didn’t want to fool anyone into thinking AI wasn’t involved, so we left artifacts of the AI’s process and limitations in the final artwork. You’ll see the misspellings and stylizations “NigHTieIfe”, “Shoping SHOHPPING, SHOPiN”, and “FOOOOD” which we deliberately did not correct to show Midjourney’s limitations. There was a lot of intervention on our designer’s part. It wasn’t as simple as “push a button and the project is done”. Will this become a core aspect of our design process? No. Did it work for this project? We think so, but it’s a very particular project where we get a lot of creative leeway and experimentation makes sense given the client and context.

After the below images, you’ll find more on the process as well as a statement from the artist.

arts and nightlife cover art showing a collage of ai-generated figures such as musicians and a drag queen
shopinig shohpping shopin cover art with a collage of ai image generated eyegasses, shoes, a dress, shopping baskets, a wine bottle, grocery store aisles, and vending tents
city living cover art with an ai generated cityscape, and illustrations of a smiling dog, a smiling cat, four people walking backwards on their hands and feet, a hairdryer, dumbells, and a frowning woman
fooood and drink cover art showing ai generated images of pancakes, sushi, a cucumber, various cocktails, a cut of meat, and a mushroom

Statement from the Artist

As an artist and designer, yes, these tools of ersatz creativity are threatening. As a technology geek they’re exciting! As a human being, the potential for destruction and misinformation via fabricated images and videos used to create documentation of things that didn’t happen–to or with people that didn’t actually do–them is terrifying. Photographic evidence of a Democrat cabal eating babies, anyone? But the technology is here and it’s not going away. So I decided to collaborate with it. I asked the Midjourney bot and DALL•E 2 to create lots of little elements using written prompts that I created. Then I touched them up and/or collaged them together to create these final compositions. Will I use this method all the time? Probably not but it was a fun experiment. I can definitely see purchasing stock photography as a thing of the past. It might also put me out of business and I don’t know what to do about that besides try to make it work for me, and not against me. 

When discussing this project with Marke, we landed on using a woodblock print look to reference protest art and a means of production that is low-tech and accessible to anyone with basic materials. However, I wasn’t sure how to achieve a woodblock print look digitally. Creating a bunch of woodblock prints wasn’t really feasible. I ran across Midjourney and this seemed like a really interesting solution to experiment with. 

Something I love about this technology is its quirks and artifacts. The images in this Best of the Bay art are very specific but pretty crude. However, when you ask these things to create more complex images with lots of elements, on first glance, they look completely convincing. But zoom in on people in the distance or an intricate piece of equipment and you begin to realize that these images are really nothing at all. I think the Food & Drink art is a good example of this. If you really look at the stack of pancakes or piece of sushi … what ARE those things?

–Aaron Joseph, Artist & Designer


Our Process

We also thought it added interest to the work. We asked the AI to create a number of items in one image such as musicians, food items, pets, or miscellaneous objects. It would give us back 4 options, and we’d pick the one that was most recognizable or interesting. It would generate 4 more images, each containing a number of items. Some of those results were completely unrecognizable gibberish-as-drawings, some looked vaguely familiar. We also asked it to create star burst graphics and medallions to use as graphic elements. Text was particularly challenging for the AI tool, which we think ended up creating really interesting results. The text you see is each a collage of many different attempts at getting Midjourney to create written text as a drawing, which it seemingly wasn’t designed to do. We then cut out our favorite items, collaged them together, added color and texture, and came up with final pieces you see below.