Unleash Your Creativity!
An Introduction to AI Art Generation
It’s been swirling around out there in the metaverse for a brief moment, and suddenly, just like that, AI has become a standard part of our everyday software. In particular, Generative AI drawing features are now included in Photoshop, Illustrator, InDesign, Storyline, Vyond, and several other drawing and content creation tools.
At first, the drawing capabilities were a bit crude, with AI-generated artwork containing strange artifacts such as misshapen hands, faces, and abnormal limb counts. However, over the past year, this has greatly improved, and now AI artwork is a viable contender for assisting with the quick creation of engaging learning content.
A Recent Example
I had to create an infographic about the spread of malaria, and the project required images of a bird, a mosquito, a horse, and a human. It would have been easy enough to go to our stock image provider to find these, but locating images that matched color and visual style would have taken some time. A quick search would return all manner of images: JPGs, vector, photo-realistic, cartoonish, illustrated, icons, etc. You get the idea — a hodgepodge of styles.
In this case, there was a simpler solution: I opened InDesign, drew a circle (InDesign lets you create AI art in whatever frame shape you want) and entered the prompt for each of these images. In mere seconds, I had beautiful images all sized correctly and in a consistent visual style. See the examples below.
These images would have taken much longer to crop and format if I had used a stock image provider, and drawing them from scratch would have taken even longer still. Score one for AI.
So, How Does it Work?
Let’s take Photoshop as an example. There is now a dialogue box that appears when you have a new artboard selected, allowing you to describe the image you would like. You can choose an art style, photorealistic style, or even upload to a sample image you’d like to use as a suggested style. See below.
The tool is really good at understanding what you’re trying to convey, and the more detail you provide, the better the results. Let’s type in an example and see what we can get. How about: “Person standing on city street holding a cup of coffee and looking at their phone.” Here’s the result:
The woman on the left was done in Artistic style, and the man on the right was done in Photographic style. I think both look pretty good!
When I entered the prompt, Photoshop returned three options to choose from. Note that a high level of diversity is embedded in these programs in regard to race and skin color, but you can also clarify that if need be. Below you can see one of the other choices it rendered for me. Notice her hand holding the phone; the little finger looks very odd, almost like it is going through the phone, and the other finger seems extremely long too. Most people would likely miss this with a quick glance at the image, but you need to carefully scrutinize everything to make sure and catch issues like these.
Another drawback is that if you really like this woman and her outfit, but want her holding a bottle of water instead, most generative AI tools don’t easily carry elements over. There’s no way to say, “Use this image, but change this…” Instead, you must create a new prompt: “Woman with dark hair wearing red shirt and grey blazer standing on a city street holding a bottle of water and looking at her phone.” You can see below what was returned:
Not a bad image, but it’ definitely not the same person as in the previous image. You could also argue that some of the fingers still look a bit unrealistic, but I think overall, you’re beginning to see just how powerful and useful these tools can be.
When used in the right context, AI-generated artwork can really enhance and speed up your workflow. When requesting AI images in the learning environment, instructional designers must be as clear, detailed, and descriptive as possible, and remain flexible when changes are required because right now AI generation always makes a fresh new image rather than duplicating and modifying parts of an earlier one.
SO, How is AI Art Made?
Without getting too technical, I’d like to briefly touch upon how AI art is created. The first step in the process involves NLMs, or Natural Language Models. The computer is trained to recognize what each word means, what it infers, what it’s similar to, and what it represents. So, when a person enters a prompt, these words get ‘translated’ into computer speak, which then puts together an image that reflects the request. For example, if you type in, “Show a person holding an umbrella,” the computer can infer that it’s probably raining, and that they shouldn’t be holding the umbrella the same way they would hold a baby. Imagine a vast network of nodes that connect in all directions; that will give you an idea of how the computer ‘thinks’ about the current request in the background.
The next interesting part of AI art is called the GAN, or Generative Adversarial Network. The AI drawing process can actually be broken down into two distinct tasks. The first task is to create images based on the prompt (this part is called the Generator), and the second task is to critique or compare these images to see whether or not they look real. This second part is done by a program that has been “trained” on thousands of images and has learned to determine how smooth an apple should be, what color a sunset is, how many tires a car should have, or what perfect latte art looks like (this part is called the Discriminator). This process repeats over and over until the generated images are approved and then sent to you.
There are some other processes involving techniques like Diffusion and Neural Transfers that help generate images matching a particular style, but I’ll keep this short.
Currently, the art created within these drawing tools is copyright-free and can be used commercially. However, there’s fine print that specifies if you create offensive or otherwise NSFW (Not Safe for Work) artwork or use copyrighted material for your prompt, you may lose your privilege to use the tool (at least that’s what Adobe currently indicates). It’s probably safe to assume that whatever you enter as a prompt is being reviewed by their team to help train the software and prevent misuse, so keep that in mind. Additionally, the AI tool itself may flat-out refuse to draw something controversial, confusing, or overly specific. For example, ask for a portrait of a famous person and you probably won’t get what you want. And one more thing, images you make with AI are starting to be tagged by online platforms with a little note that reads, ‘Generated by AI’ or a CR for ‘Content Credentials’ or something similar to let viewers know how these images originated.
All in all, I think you can begin to see the power and creativity these new AI features can bring to designers and media developers. Check out this recent blog example by my coworker Walter Schirmacher where I used AI to generate the perfect image. And just in case you were wondering, yes… I did have to remove an extra finger from the monkey’s hand.
So anyway, the next time you get a project where this may be the right solution, try out a prompt and see for yourself how it works for you. What an exciting time to be a creator… am I right?
______________________________
Additional Resources