A few weeks ago I ran into two different groups of javelinas — wild pigs that aren’t actually pigs — at Bosque del Apache National Wildlife Refuge.1 I was completely enchanted and tried to make some AI versions of them. I created a prompt with a single word, “javelina,” and forgot to specify a rectangle instead of a square. This was one of the very first results.
I liked the stylized nature of the image even though the critter’s snout was way longer than a real javelina’s. So I plugged it into Midjourney’s “describe” function, which creates word prompts based on an uploaded image.2 It coughed up some rather wordy prompts that I didn’t fully understand, most of which used the word “wildlife.” I tried one of them and … voilà! A deer showed up.
The image above is somewhat similar to that of the javelina so it seems the “describe” function actually worked pretty well. I tried another of the prompts the function had suggested and Midjourney came up with the image below … which I like quite a bit.
Last, Midjourney has made it easier recently to see and copy other folks’ prompts.3 The image below comes from one of those prompts, which I modified for my own purposes.
I posted photos of some of the javelinas I saw, along with a few other critters, on my other Substack.
“Describe” is an indirect — not to mention imperfect — way to put an image into words, then back into an image again.
In case you’re worrying that copying a prompt is copyright infringement, it’s not; the US Copyright Office has ruled that AI images can’t be copyrighted. Not only that, traditional artists claim AI image generators are, by their very nature, infringing on their copyrights. And then again, my own prompts are also visible to — and easily copied by — other folks.
AI ethics are extremely unclear at the moment (a major understatement). At a personal level, I wouldn’t use someone else’s AI-generated image and pass it off as my own, even though it would be perfectly legal to do so.