This week for MCO 426, we studied further into the world of copyright, and generative AI’s place in it. For this blog post, I was tasked with comparing three images: one I’ve taken myself, a creative commons public domain image that resembles my photo, and a generated replica of said images using an AI image generator.

Let’s play two truths and a lie:

Image 1 is a real picture of the moon.
Image 2 is a real picture of the moon.
Image 3 is a real picture of the moon.

It’s glaringly obvious right? So which one is the lie? If you showed your mom any of these moons I know she won’t be able to find the AI generated moon, but you might be able to if you have a trained eye for AI content like I do. Honestly, this one is tough it could really be any of them. Image 1 is a bit blurry which we see in less popular image generators. Image 2 has way too many details which more mainstream AI models tend to do. And the third one could pass for a real image… If it was one!

I actually kept the order of each moon. The first image was taken by myself with my own telescope and phone camera. It took me several attempts to get the image you see, without it falling out of view of the eyepiece. The second image is actually produced by NASA which explains why it is super detailed. The third image was produced with ChatGPT which I believe uses Dall-E. I first tried to use the actual Dall-E4o GPT button but it instead used an older model and the image it made was pretty bad for moon standards:

Swapping back to regular ChatGPT4o I gave it this prompt:

“generate a realistic image of the moon taken from a telescope. its a half moon, the shadow of the moon fades over the surface creating extra contrast on its craters. left side of the moon is the illuminated side”

And honestly it doesn’t look too bad. If you weren’t paying too close attention that is one moon alright! But the thing is, someone is going to mistake it for real. If there ever was a time to question the authenticity of digital media, it is now since AI is blurring the lines.

As AI generated content becomes more realistic, it raises some serious questions. If an AI can mimic the look and feel of astrophotography, who really owns the image? For the large language model to produce an image of the moon, it must be trained off of hundreds or even thousands of images of the moon. Does that make the AI moon derivative off of public domain images? Is it stealing if the LLM was trained off of copyrighted images? What if a generated image is nearly 100% identical to a copyrighted image? Does that make it a copy, and is it stealing?

We already live in a world where AI is used to make fake but believable media for portfolios, websites, advertisements, and even science projects which is cause for alarm. Then there’s creators and hobbyists (and amateurs like me) who spend their time and effort into trying to capture the perfect moment through their own lens. How do we compete with AI if it can generate the dream image in seconds?

I for one love new technology even if I can’t get my hands on it. I think generative AI is a phenomenal tool that when used correctly is an indispensable tool for completing creative visions. But to completely replace an already established art form completely removes the integrity and what makes that art have value. There is always a human aspect behind all great art, even if its something as distant as the moon.

Image 1: Photo by Peyton LeDrew (April 2021).
Image 2: “photo of moon” by NASA via Unsplash. Free to use under the Unsplash License.
Image 3: AI-generated image created using DALL-E 4o by OpenAI, based on a prompt written by Peyton LeDrew.
Image 4: AI-generated image created using DALL-E 3 by OpenAI, based on a prompt written by Peyton LeDrew.