Our project this week was to use an AI image generator of our choice to generate three rounds of images. The theme I chose from the provided list was “taking the bus to the roller rink,” and I used Microsoft Bing Image Creator to generate my images. After writing my first prompt, I made this quick sketch of what I wanted my image to look like:
The first prompt I wrote was:
A pop art illustration of a teen girl on a city busy. She is wearing tennis shoes, cutoff jean shorts, and a colorful windbreaker. She is listening to a walkman and looking out the window. She is sitting slouched down in her seat with one foot propped up on the seat back in front of her. On the seat next to her is a color-block duffle bag and a pair of pink roller skates. Out the window of the bus is a blurry cityscape. The image is rendered mostly in aqua blue, bubblegum pink and fluorescent yellow.
With this prompt, Image Creator generated the following four images.
I was honestly pretty surprised by how close this got to my initial vision in color, style, and mood. However, these images were missing a few details, including the big hair (which I forgot to write into the prompt) and the roller skates (which I did not forget to write into the prompt). For fun, I also decided to add a different setting in the second round—in my head, she’s a fun-loving girl stuck in a stripmall town without much going on, so the roller rink is her outlet.
A pop art illustration of a teen girl with big hair on a city busy. She is wearing tennis shoes, cutoff shorts, a giant hair bow, and a colorful windbreaker. She is listening to a walkman and looking out the window at strip mall. She is sitting slouched down in her seat with one foot propped up. On the seat next to her is a color-block duffle bag. On top of the duffle bag is a pair of roller skates. The image is rendered mostly in aqua blue, bubblegum pink, and neon yellow.
In the second round, I think the AI did fine with the hair and decently well with the strip mall (even though most of the buildings literally have “STRIP MALL” written on them), but it had a lot of trouble with the roller skates.I tweaked the prompt to try to get some of the details closer to what I wanted:
A pop art illustration of a teen girl with a bored expression and big hair on a city busy. She is holding a pair of roller skates on her lap. She is wearing tennis shoes, cutoff shorts, a giant hair bow, and a colorful windbreaker. She is listening to a walkman and looking out the window at strip mall. She is slouching with one foot propped up. On the seat next to her is a color-block duffle bag. The image is rendered mostly in aqua blue, bubblegum pink, and neon yellow.
Well, at least there was a minimum of one skate in each of the images in the last round. Not important, but I thought it was interesting that instead of STRIP MALL, it’s pulled some other text (variations on “city bus”) from my prompt to put on the sides of the buildings.
This was fun and challenging. I tried to make the request even clearer in the third round, but the skates were on her feet no matter what, sometimes missing wheels or combined with the tennis shoes or suspended in the air for no reason.
As far as usefulness, these are not images I would show to a client or use for any final product. They just have too many little anomalies—and because they are generated images, they wouldn’t be very easy to tweak. Where I do see this tool being helpful is in generating ideas and helping to flesh out my original vision. The small differences in mood, color palette, style, and composition of each of the images is helpful to me in visualizing how I could take my very rough sketch and turn it into a fleshed-out final illustration.