Demistifying AI

Exploring AI-Image Generation

I started this activity with hesitation, really not excited to be creating AI art, so I must preface this by saying that the only reason I am doing this is for a grade, but also the willingess to learn, but that I really don’t see anything good about using AI to create art.

I wanted to work with my cat and create images of her and see what it could come up with. I used Stable Diffusion for this.

The first prompt I input was “calico cat” because, well, my cat is a calico cat. Here is what it spat out:

It doesn’t look like my cat, and it definitely feels more like a tabby calico than just a calico, but I knew I’d have to refine details to get closer to the image I wanted. Already, I was feeling a little weird looking at it. It has that flair that AI art has that makes me uneasy.

Next, I input “closeup of a shorthair calico cat with green eyes with a bigger build” and this is what it gave me:

It’s closer. The cat definitely looks a little more thin than my beautiful, chunky (said with love) cat. The patterns are still a little off, but I didn’t give it specifics about the pattern, but it does have mostly the right idea. I like the detail it has in some areas, but then there are other areas that just look airbrushed in and it makes me feel weird.

Next thing I told the AI to generate was “shorthair calico cat with green eyes, a pink nose, with a bigger build in the style of Mona Lisa,” which it responded with this:

And hey, that’s starting to look like my cat! But it’s not in Mona Lisa style, but that may be my fault because I still had it set as the “photograph” setting. So here’s me retrying without any settings:

Now I’m confused, because it’s like half Mona Lisa-esc but also just not at all. So I decided to play around with some of the model settings

This is the same prompt but in the Rococo model, and it’s actually really cute, I won’t lie. I like the random bow that appeared and the colors.

I did one more just to see what it would spit out:

This is in the Baroque style. I guess, of all of them, it did get the Mona Lisa style down the most, but not entirely. I’m sure I might just be putting in the wrong queue words or not being specific enough, but I think it’s cool.

My overall thoughts on this?

It’s cool, and fun to play around with, but not cool or fun enough to ignore the threat that it poses to artists.

I think it’s really, really threatening. Even if it doesn’t take from artists like people said, how can you prove that it doesn’t? It learns from inputs and it learns from taking images and prompts and anything you can find on the Internet to build it. It learns, and it implements what it learned into its model for future purposes. It feels relentless, like publishing any art on the Internet puts you at risk of having it taken and tweaked just so someone who knows how to type in a good prompt can get the same output. There isn’t creativity in prompt-making, but there is creativity in creating art.

It’s not art, it’s data, but the data is built off of previous ideas of artists, taking it and transforming it into some code to create more data.

I don’t like it.