There were a few things that I generated that ultimately didn't get used for my Evolution by AI project. For example, I first tried drawing the starting microbes using different models of Stable Diffusion. One of them gave me these:
I liked these, but they had a certain sameness that I thought would make the project less interesting. In other words, I thought that the variate tool wouldn't have enough to work with to make successive generations distinct enough to produce unique clades.
Another model gave me these:
If anything, these are even more uniform in appearance. Even so, I decided to do a quick test with this model. I took a single image and ran Variate on it. Then I ran Variate on that one, etc., for 10 generations. Here they are, from left to right, top to bottom:
As I feared, the Variate tool wasn't finding much to vary.
Fortunately, Stable Diffusion 1.5 gave me a set of starting images that were visually appealing and different enough that Variate had something to work with.
Even then, my first attempted crashed early. I only made it through two generations before I made a mistake and had to start over. Here are the first two generations of that failed first attempt (only the fourth image survived to the next generation):
Generation 1
Generation 2
I was on tenterhooks for the entirety of the second attempt because I thought I might make a mistake, again. I did make one mistake, but I was able to recover from it without imperiling the rest of the experiment.
These illustrations were drawn using Stable Diffusion 2.1.
