There is a lot of talk and controversy about AI and its means of generating images. It almost certainly uses images on the Web but does it steal intellectual property or is it derivative in terms of fair use?
First off, here is how Google Gemini says it works, based on what it found on the Web.
“AI generates images by training on massive datasets of images and corresponding text, learning to associate visual patterns with descriptions. Using neural networks—commonly via diffusion models—the AI starts with random noise and gradually refines it into a detailed image that matches a user’s text prompt. These models understand, rather than copy, styles, shapes, and colors to create entirely new, unseen visuals. “
Further, it steps through how…
Core Concepts and Techniques
- Training Data: Models like Stable Diffusion, DALL-E, and Midjourney are fed millions of image-caption pairs to understand the relationship between text and pixel patterns.
- Diffusion Models: The dominant method where the AI learns to reverse a process of breaking down images into noise. It starts with, or introduces, random pixel noise and iteratively removes it to reveal an image based on the prompt.
- Latent Space: The AI maps concepts, styles, and objects into a high-dimensional mathematical space, allowing it to translate descriptive text into specific pixel arrangements.
- Generative Adversarial Networks (GANs): An older, though still relevant, approach where two neural networks compete—one generates an image, and the other evaluates it for authenticity.
In short, it’s a lot of mumbo-jumbo that you may or may not understand. I certainly don’t!
I put it to the test by asking it to generate a picture of Gary Gygax in the style of the Mona Lisa painting. It came up initially with Mona Lisa with a beard. I told it Gary had little hair so it kept her hair but he was now bald on top. Then I said, remove Mona Lisa’s hair. Now we had Gary in Drag. I said put him in masculine cloths. The final result is thus….

For the comparison, here is the original painting…

The style is definitely there. However, it does seem that the new “painting” is nothing more than a collection of images assembled seamlessly into one image without attribution to the original artists. That is required for fair use. The newly generated picture does not actually look like Mr. Gygax but does bear many features. Interesting painting and AI did kind of nail the vibe I was going for. I did not care about the background other than it being similar to the original Mona Lisa. The books, dragon, volcano and so forth were nice touches.
So, am I a convert? No. I think there needs to be a lot more thought put into image generation with respect to IP law. This does clearly violate fair use practices as there is no attribution to the original artists. It’s an interesting experiment.
AI generated content is starting to work its way into commercial gaming material. Before buying, you should consider whether or not the items in the documents/books have been properly attributed to their respective artists. For instance, if it is AI generated from my own works collected in my own library, that would be perfectly fine. However, this almost certainly is not the case with 99% of AI generated material out there. I suppose you could stick with free to use sources on the internet – those sources that are old and no longer bound by copyright law or those that the artist/author grants express permission. AI, however, is prone to make mistakes. So even those instances should be looked at with a dubious eye.
So, if/when I ever publish anything will I ever use AI for my artwork? No. I am not interested in walking down easy street. I prefer to take pride in everything I do and own the mistakes when things go wrong. It’s who I am.
Posted by John 

