It seems like more and more people are using Artificial Intelligence to generate images, now that it is accessible to anyone with an internet connection. On the positive side, we can think of AI as another tool (or medium) for artists to use. On the negative side, anyone with a computer is able to create an image that can be called "art," minus the messy old paints.
Granted, the AI-created image probably couldn't pass as "real"—yet—but it may be just a matter of time. The computer has created "paintings" that have been sold at auction for the price of a new car. Or ten. And art exhibits have needed to create a new category to display computer created images, so that the viewer knows exactly what they're seeing.
I started my introduction to AI with Deep Dream Generator. Initially it's free to try, but eventually I subscribed ($19 per month, cancel any time) for more usage. I also tried StarryAI and Drezgo, but of these so far, I like the way Deep Dream thinks.
Then I got to work. Or play. At the time, I was working on a painting in my life drawing group. This is the photo of our model, Michael, that I used for reference:
As a character, Michael looks like he could be swinging on vines through the jungle, slaying Vikings, or posing for a Romance novel, but I wanted to give my painting a fantasy look. I like to use oil paint for my non-illustration work, and most of my oils have a surreal quality, so the fantasy direction wasn't so far off. Here's what my unfinished painting looked like at the time:
Note that I didn't paint the dragons you see in the painting. I found some online, and pasted a couple of them into the painting (in Photoshop) for ideas regarding placement, size, color, etc.
At this point in my learning curve I could use one of three ways to generate an AI image to affect my painting. 1) I could try a text prompt only, without giving it a base image to start with. 2) I could scan the photograph of Michael into an AI program to use as a base. 3) I could scan in the painting with the dragons as a base. I was curious to use all three to see how AI works.
The first test was without using a base picture. I just gave a text prompt: "Man in forest looking at dragons."
Next you can add modifiers before generating an image, like style, influence, lighting, genre, etc., that are listed for your selection.
I chose this combination:
Artist: In the style of Greg Rutlowski (a fantasy artist)
Quality: photorealistic
Effects: cinematic lighting
This is what the computer gave me:
Hey, That was easy. I feel like an artist already! it's a nice piece of art. Not exactly what I had in mind, but I can certainly see it used for a science-fiction book cover. And it only took two minutes to create. It probably would have taken me a few days to paint this image, if I even thought of it. The AI program pieced this together from images taken from the web, so I can't be sure that parts of it would be legally unusable, commercially. Rule #1: Avoid copyright infringement! Any image you take from the web is property of the artist. You can't just steal it for your own commercial work.
So do we like AI yet? How can illustrators use it as a tool?
Once the user learns the limitations of the program, they can, for the most part, control the direction they want an image to go in. The possibilities are endless. AI computer programs analyze thousands of images in cyberspace and combine them to create new ones as specified by the user. Sort of. This is where the skill of the user really counts. More experience will lead to more a more desired direction.
StarryAI has some nice examples that people (or their computers) have made from text prompts like this one very long prompt :
Symmetry, Beautiful Cat,Germanic Vestal Sacralpretty Face, Intricate, Elegant, Highly Detailed, Digital Painting, Artstation, Concept Art, Smooth, Sharp Focus, Illustration, Art By Artgerm And Greg Rutkowski And Alphonse Mucha:
I only wonder how many images this artist generated before getting this one!
Someone else tried a prompt that's simple: Robot in a suit:
The artists have made these images public, so I'm sharing their work here (not for commercial use) without their permission.
The point is that you can start with a text prompt of anything or combination of things. And every time you generate, it will look different, even with the same settings. You never know where it's going.
For the second test I scanned the original photo of Michael into DeepDream so it can generate from a base picture. My text prompt was "Man with dragon wings, looking at flying fire-breathing baby dragons.."
I chose these modifiers:
Artist: In the style of Greg Lutkowski
Quality: photorealistic
Effects: cinematic lighting
Then the most important part is to determine how much I want the photo to be manipulated. Do I want it only slightly changed, or greatly changed? I set it for an 80% change, and this is what the computer created.
Wow. Admittedly, it's better than anything I could paint! But I guess an 80% change is really off the chart compared to the base art. But I can see how an illustrator may be tempted to go this route. It's fun. It's easy. And it's probably plagiarizing Mr. Rutlowski's work.
I tried again with the photo, this time with the same modifiers, but setting it for 20%. It's interesting how the computer saw the blue tape on the model as objects. (We use tape to get the model back in position after breaks.)
This is 40%:
Yikes. From here I can keep starting over, or I can use my latest image to "evolve," meaning the computer will use this as a base and take it further with each generation. I went through many evolutions and generations, so I don't remember what modifiers I used or what percentages, but AI gave me some interesting images:
I stopped playing with that particular "evolve" when Michael started looking more like a Michelle. But I do love this image and can see it as a painting. (Except for the creepy hands!)
Next I scanned in my painting with the Photoshopped dragons to use as a base. More interesting stuff:
40% in the style of Greg Kutlowski:
60% in the style of Alex Alemany, another fantasy artist:
50% In the style of John Jame Audubon:
&0% in the style of Wes Anderson. Are we having fun yet?
Now I suppose you want to know where I ended up. Well I ended up crossed-eyed and overwhelmed, but—spoiler alert—once you start going down that rabbit hole, it's hard to get out.
OK- back to my painting. Remember, I wanted to use AI as a tool for ideas—basically to see something as I wouldn't ordinarily envision it. After searching through all the generated images (about 50 of them!), I decided that I liked the background in this one:
The color palette and style of the hazy forest was nicer than what I had painted from my own imagination.
So this is what my painting looks like now, still not done, but painting the background in the style of the above pic was a good decision. I also like the dirt all over the AI guy (and Michelle), and will add it to the model. After all, if a guy is sitting in the woods, contemplating dragons, he'd probably not bathe regularly. And I wouldn't have thought of that if I hadn't found AI. Sorry Michael. Get ready for the mud pit.
I'll post the final painting when done. Meanwhile, bear in mind that my experience will not help you much in your exploration of AI, as I barely touched the surface. You''ll have to start from scratch, too. Google "AI image generators," and try some that offer free trials. And watch out for that rabbit hole. It's a time killer...
Update: Here's the final painting:
Comments