🖼️ Having some fun with AI Image Generation
Since I’m messing around with the Language Model side of AI, I decided to have a look at the Image Generation aspect. Not going to lie, it’s pretty amazing.
You type in a few keywords and a couple seconds later out pops a picture. That alone is impressive, but it’s not some copy/paste job, it’s a composed image. And there are some amazing renders out there, like Cloud cities, representations of “artificial intelligence,” and some surrealist paintings that have an incredible level of detail. And let’s be honest, there are a whole of lot of nudes out there. The machines are being taught some unusual body proportions.
Not every image is a gem, but considering what’s going on in the background of taking different kinds of images and layering them together, it’s pretty impressive. Even the worst image is better than my best attempt. Of course, if you don’t like what you get, spin the wheel and try again. A few seconds later a new image will appear. That’s something to marvel at. There is a lot of fun in playing “What if” or “What would this look like?”
That leads into the language of AI, the right keywords in the right order, to generate an approximation of what you asked for. The AI is quite good at generating a background of a beach with palm trees in it. It’s also has a solid understanding of the Steampunk theme. Asking AI to draw X, and add Steampunk to the keywords and you get that rich leather, brass, and copper feel.
It’s pretty incredible how the image comes together. Images that represent what you’ve asked for are layered and blended together to make it look like it was all done as one piece. A far cry from the hack job of pasting web images into Affinity Photo.
So, you have to ask yourself, is it art?
Sure, why not? Why wouldn’t it be? It’s no different from fractal art, kaleidoscope images, or mandala. Those look fantastic, a beautiful blend of color, shape and design, and computer generated. AI art is the same thing. It takes trial and error to get the right keywords. Someone has to dream up the art concept and go through the process of creating a usable image. And it can be a beautiful blend of color and design.
Raccoons on a beach may not get any likes, but some of those floating cities are amazing and should totally be printed on canvas. Is it the same as doing it by hand? No, but it’s still pleasing, so I’d say it was art.
Are there AI artists? Again, why not? It’s not the same as the hours it takes to master light and color using oil, but there is still a craft to generating a decent AI image. You have to know the style to ask for, the keywords, a level of creativity and imagination. It’s not DaVinci, but it’s not a scribble either. Maybe there is an art to using the right keywords to express the right concept? Will that be a thing in the future, AI artist for some company?
Regardless, it’s incredibly fun. You can truly let your imagination run wild and dream up all sorts of things. I can’t draw anything beyond stick figures, but with a few keywords, I can generate an image that expresses what I’m thinking. And in a really close approximation. I have no artistic talent, but I can create art for my own amusement and enjoyment.
Plus, you can blend things together in new ways. Is there such a thing as Steampunk Spider-Man? What does it look like if you put character X in location Y? What does it look like if I blend these genres together?
Equally, you can make reference material for your own creative process, like generating a medieval tavern scene, a character dressed a certain way so you can reference it later, or a certain type of night scene. You can immerse yourself in the writing process.
Need help conjuring up the right look for a villain, let AI cook one up for you.
Need some help with the visuals of an airship, let AI lay it out for you.
Need some help visualizing the creepy entrance to a dungeon, let AI sculpt that for you.
That is a pretty powerful tool to have at your disposal.
They can’t all be gems though. While AI is pretty bad ass in what it generates, it’s not hard for it to go off the rails. Don’t be surprised if you get 3 arms, 8 fingers on one hand, or all manner of anomalies in the background. A Three-armed tavern keeper might be a cool super power, or it maybe not, but it can happen.
AI is good, but it doesn’t always connect the lines. Pieces go missing and random elements can appear. Some times it’s quite comical, other times it’s just annoying. It gets 99% of the image correct, and looks great, then you have this totally mangled element thrown in. When AI gets it right, it’s looks great, when it goes wrong, it goes a bit Skyrim.
Even with the glitches and some of the goofy things people are doing, AI Image Generation is impressive. We’re now at point where you can type in some words and get an image. And it’s accurate. It’s still in it’s infancy, and there are glitches, but I think it’s awesome.