AI-Generated Music Images are Really Bizarre

Like many, I have been playing with DALL-E3 - from Open AI - for quite some time now. While I see images created by this pretty amazing AI model on websites LinkedIn a LOT, I really only use it for AI demonstrations. OpenAI developed DALL-E, which was first released in January 2021. The name DALL-E is a combination of the Spanish surreal artist Salvador Dalí and the Disney robot Wall-E. DALL-E3 can generate images in a variety of styles, even for concepts that don't exist in the real world. It can also imitate different artistic styles and works of real humans. In my experience, the images that DALL-E3 creates for music education based prompts are often riddled with mistakes and can even be pretty problematic. For example, take a good look at the image above. The prompt that I typed to generate this image was “Create an image of a middle school string orchestra rehearsing being conducted by a robot.” Notice anything? To me, it appears that every student in the image is of Asian descent. I didn’t specify the race of the students at all and yet this is what I got. The stereotype that Asian students play string instruments is not only wrong, it’s pretty offensive. Aside from that aspect, the image is riddled with bizarre things, including: they are playing their instruments with the wrong side of the bow, some of the bows are curved, some instruments are missing, and the music on the board is pretty funny. I think that if you are looking to have some fun with AI, and to show students that AI isn’t quite “there” yet, you should actively solicit prompts from your students and then find all of the anomalies in each image. It is a perfect “fun” activity if you’re either discussing AI or you have a few minutes to spare at the end of a class. Here are a few images complete with the prompts I entered. See if you can find what’s wrong!

I just entered the same prompt (“Create an image of a middle school string orchestra rehearsing.”) and got an image of a diverse set of students playing string instruments. But when I zoomed in, I found some FUNNY things. Look at the image above. What’s up with their faces? Some of the kids are missing the lower half of their body. There seem to be lots of chair legs without seats and backwards stands. Weird. Here’s another one:

This is the prompt for this image: “Can you create an image of a high school girl practicing the french horn?” Notice anything? First, she is missing her right hand. Second, she has at least 6 fingers. I don’t know about you, but I don’t know what Rusic or Rection is. I would imagine that my students would be giggling a bit if they saw that. There are also some alien musical instruments on the shelves.

For this image I entered the prompt “Can you create an image of the John Philip Sousa marching band performing in the early 1900s?” Pretty decent job but what the heck is the guy in the bottom right corner playing? How about the headless musician playing the sousaphone on the left? I also love the floating bass drum in the bottom left corner.

The prompt for this image is “Can you create a fingering chart for the flute?” This is what it spit out. A good try but what is going on here? That isn’t an accurate depiction of the keys on a flute, the last time I checked the keys are labeled with alien numbering and the actual fingering chart is borderline surrealism.

I think this one is really funny. I entered ”Can you create an image of a man performing the oboe at a recital?” This is what I got. First of all - what on Earth is he playing, and why does it look like he is simply holding the mouthpiece up to his cheek and blowing it like a flute? I love that some of the audience members are holding what appear to be melting cellos in their laps. The piano looks pretty accuarte but I feel sorry for the pianist who has no eyes.

Talk about weird. The prompt for this image was “Can you please create an image of a tenor singing on stage in the role of Papgeno?” Love the GIGANTIC panpipes. What are those Scottish dudes playing in the background? Was this image taken at a recent production of The Lion King???

I entered “Can you create an image of an elementary school chorus performing at a holiday concert?” At first I thought, Wow - finally a relatively problem-free image. Until I zoomed in. These kids look like they are all half rodent when you look at their faces. I would think that this chorus would scare the daylights out of any audience. Eek.

I am quite certain that DALL-E3 will improve over time and images like this will be a speed bump on the progress of Generative AI. But until then, I am going to keep prompting and keep giggling. I hope you and your students enjoy!