Illustrating a silly song…
12 minute read
March 25, 2025, 2:30 PM
So after two “downer” Journal entries in a row, I pledged on social media to make the next entry a fun one. While I was operating the train and running various Today’s Special songs through my head as I did, I came up with something: the “Blue Cow” song by Clive and the Cowboys. That one goes like this, singing about various silly things along the way. I’d been wanting to do some humorous illustrations with an AI engine for a while, and this seemed like a perfect opportunity to do that. This time, I used Meta AI, largely because they seem to do a better job with illustrations than Bing (which I have played with before), plus Meta, unlike Bing, doesn’t throttle you after fifteen inquiries.
In doing this, I tried to keep as close to the original lyrics as possible, deviating only if the original lyrics either wouldn’t make sense as a query, or if they produced weird results and I had to refine. As I go through this, I’m putting the lyrics underneath the photos, and then if you click the photos, you’ll see the AI query that I used. For Meta, to get it to do illustrations, you start your query with “Imagine” and then give the description of what you want.
So here’s the result.
Blue! Blue! The cow was blue! She jumped high like a kangaroo…
The tree went “baa”…
And the car, it flew, and this silly, silly song I will sing for you.
Two beavers playing catch…
The porcupine gets scratched…
The crocodile was knitting two socks that didn’t match…
The tree was trying to dance…
And the fence was planting plants…
And a big, tall giraffe wore a pair of brand new pants!
Blue! Blue! The cow was blue! She jumped high like a kangaroo…
The tree went “baa”…
And the car, it flew, and this silly, silly song I will sing for you.
The sun was out at night…
The turtle flew a kite…
The aeroplane went swimming, that was a funny sight…
Ha ha!
The hippo ran a store…
And the door started to snore…
And playing on the bagpipes was a purple dinosaur!
Blue! Blue! The cow was blue! She jumped high like a kangaroo…
The tree went “baa”…
And the car, it flew, and this silly, silly song I will sing for you.
And that’s the song. I was amused doing this because, as we’ve seen plenty of other times before, generative AI is not the best in putting things together to make what we intended, even with the best of descriptions. While the descriptions of the cow did more or less what I had intended, with the only tweaking of the query’s being to give Emily some extra height on that jump (after all, we have standards), some of the other ones produced some downright strange results and required some additional finagling in order to get what I wanted.
The first venture into the absurd was the line where the tree went “baa”:
Going into that, I wasn’t sure what I was trying to go for with the tree’s saying “baa”, but I ultimately went with a speech balloon. I certainly wasn’t expecting a sheep’s head embedded in the tree, nor did I expect those little gremlin-like characters as a more anthropomorphosed version of the tree.
The flying car went through a few evolutions in order to get the result I wanted. First, the choice of car model, i.e. a Honda HR-V, a Kia Soul, and a Model T, was deliberate. Two of my past cars, plus Gertrude. The wings were a deliberate decision, because without them, it didn’t look right.
With these, I felt like these HR-Vs were plummeting down to earth rather than flying, plus the doors looked a tad weird. Also note that this is the earlier generation of the HR-V. For some reason, all of the generative AI tools that I’ve tried lock onto that version of the HR-V when you say that, no matter how much persuading I give it to show the newer model. So I tried to get it to produce the new HR-V instead of the old one by specifying the Honda ZR-V, which is how Honda markets my current car outside of North America and China:
It was quite clear that I wasn’t going to get the new HR-V out of this no matter what I did, because it just plain didn’t know about it. Oh, well.
The next one that was a challenge was the crocodile that was knitting two socks that didn’t match. More often than not, the AI either showed matching socks, or it showed one single sock. The first iteration, where I said “Imagine a crocodile knitting two socks that didn’t match,” produced this:
Giving it “Imagine a crocodile knitting two socks that don’t match each other” produced similar results:
Though let’s admit: the ones where the crocodile is wearing glasses are absolutely adorable, even if the system didn’t fully understand the assignment.
When I ran the one of the tree that was trying to dance, the original prompt produced naturalistic-looking trees with varying levels of festivity. However, I intended something more literal, and so I threw “anthropomorphic” into the prompt in order to make it more human-like:
I almost ran that second image, but I decided that was too weird. I also had to go the “anthropomorphic” route with the fence, because when I put “Imagine a fence planting plants,” it produced both a human gardener tending to plants with a fence in the background, and a fence simply surrounded by plants. That was boring and incorrect.
When it came to the giraffe, I never thought that there could be so many different interpretations of how a giraffe might wear pants. We’ve all seen the meme about animals in pants, but this went to the ridiculous. Here’s some of the first run, where I ran the line exactly:
One thing that I noticed a lot was that it would produce coverings for the outside of the legs, such as in the third image, but not something that one would call “pants”. And then that three-legged giraffe in the fourth image just sort of has a tiny pair of shorts hanging on one leg.
When I simplified it to “Imagine a giraffe wearing pants,” I got better results:
That produced what I ultimately went with, which looks more like leggings, but it managed to get more coverage (though a lot of obvious non-pants looks along with it). I don’t know if I’d call that third image “pants”, though. I think it looks more like a jumpsuit, personally.
For “the sun was out at night”, I had to be very specific about what I wanted. When I said, “Imagine the sun shining through the night,” I got a bunch of sunsets. That’s not what I intended. “Imagine the sun’s being out despite its being nighttime” also produced sunsets. The final result that I ran with wasn’t ideal, but it worked well enough. The final prompt, “Imagine the sun in the sky at night with the moon and the stars,” produced a couple of images that I liked, but thought were too fanciful for what I was going for:
I feel like the second image is closer to what the sun might look like if it were possible for it to be out at night, but they were a tad too fanciful for things.
It understood “Imagine a turtle flying a kite” for the most part:
It really liked doing that aviator’s hat for some reason. Though one has to question what it was thinking with that hot air balloon in the third image. Also, turtles apparently don’t smile, because the only one that showed a happy looking turtle was that second one. The rest all looked either neutral or sad. Even with the one that I ran, the expression seems to say, “Yep, here I am flying a kite. I’m doing this because I always do this. I don’t know why I do this, but I do.”
Meanwhile, the swimming airplane produced a lot of unsatisfactory results. The first run, where I said, “Imagine an airplane going swimming,” produced things that looked closer to aviation accidents than something funny:
However, that original prompt also produced the image that I went with, so take that for what it’s worth.
I also tried specifying a swimming pool to make it look less like an aviation accident, and some of them were interesting:
The first one showed action, but it was far too big. The second one, I don’t know what to say about it, with skewed wings and the tail on the side. The third one was cute, but it looked like it was placed there, and seemed static. The fourth one was good, but I liked the other one with the bystanders better.
The hippo that ran the store was pretty straightforward, and all of the images that the AI produced were in the same vein as the one that I used, with only variations in the color of the apron and the products being sold. But the door that snored really confounded the AI. I had an idea about what I wanted based on the episode, but the AI didn’t produce that, but produced some downright weird things instead:
The first one was kind of what I was going after, but it looked more like a door that was sick in bed than a door that was simply asleep and snoring. The other three, meanwhile, are straight up nightmare fuel, especially that last one. I feel like those scenes would not be out of place in a horror movie of some sort.
So I substituted “sleeping” for “snoring”, and got better results:
I considered running the first one, but ended up going with the other one just because I thought it was slightly better. The others, meanwhile, show a door off of its hinges, laying on the ground. I guess the door is sleeping? By that logic, I suppose that this door and this door from the former JCPenney in Staunton Mall are also sleeping. Sure, why not.
And lastly, the purple dinosaur amused me. I was able to go with something pretty close to the original line, telling it, “Imagine a purple dinosaur playing the bagpipes,” and it produced decent results:
I saw that first one, and I realized that I was basically asking it to produce Barney, but fortunately, that’s not what happened. But, yes, for the most part, the AI understood the assignment, with varying levels of Scottishness included.
However, it also produced a surprising amount of photos where the dinosaur itself was the bagpipe, which I found a tad disturbing:
It looks like someone forcibly jabbed the pipes into the dinosaur’s back, or shot it with the pipes from a distance. Either way, that’s pretty weird. Though not as weird as this honorable mention:
So… its head is on the end of its tail, and the bagpipe is where its head should be, and its arms are holding one of the pipes while it’s got one of the other pipes in its mouth. Sure. We’ll go with that. Go figure. Technically, I suppose that it does meet the request, but that doesn’t make it any less bizarre.
Making things with AI engines has become an amusing little time-waster for me as of late when I don’t want to play a game, but I need a mental break all the same. It’s a fun challenge to get it to produce what I want it to produce through trial and error, sending the prompt, seeing what I get back, refining it accordingly, and then going again. This entry began as one of those image sessions, though I quickly decided to run with it as a Journal entry and do the whole song. Either way, I had a good time making this, and I hope that you are similarly amused.
Categories: Artificial intelligence, Today's Special
Leave a Reply