Red Oil Bursts no.1 (Generative AI Experiments)
Created //
TL;DR // Here's a first attempt at playing with generative AI tools to create images from text prompts. We're starting with abstract oil paintings.
Generative AI Prompt Variations
And... our journey into Generative AI begins. I'm starting with DALL-E. This is a generative AI model that can create images from text prompts. That means a computer is creating the image based on the text that I feed it. No person is involved.
I've played a little with some online tools in the past year since ChatGPT exploded into everyone's consciousness. But I didn't really try the image generators, as they seemed very toy-like at the time.
The thing is... improvements are happening exceedingly fast.
This is my first attempt
I'm starting simple, with a request for abstract oil painting images.
You may wonder where the prompt came from. If it's not clear, I had a weekend of fighting with AWS (Amazon Web Services) and as a distraction decided to put that frustration into a prompt.
It's a start...
Prompt 1
oil painting, abstract. draw the feeling of stress while trying to configure an aws deployment. pain. suffering. disappointment. probably lots of shades of red, yellow and orange.
I wasn't sure where to stop so I just kept going. After 15 images, I wasn't really seeing much new variation. Now, that's probably more due to the prompt than anything else.
Let's revise the prompt and see how it evolves.
Prompt 2
oil painting, abstract. draw the feeling of stress while trying to configure an aws deployment. pain. suffering. disappointment. probably lots of shades of red, yellow and orange. use interesting creative brush strokes and thick paint.
While there's some difference between the results of Prompt 2 vs Prompt 1, it's not substantial. Everything I read says there's an art to creating good prompts. I expect my next attempts will be a series of trial and error exercises.
Let's revise again...
Prompt 3
oil painting, abstract. draw the feeling of stress while trying to configure an aws deployment. pain. suffering. disappointment. probably lots of shades of red, yellow and orange. use interesting chaotic creative brush strokes and with extremely thick paint.
I removed the whole "AWS" context, asked it to be a bit more chaotic, and asked for thicker paint.
The results are still fairly similar, so I stopped after 5 images. It's odd that some tree silhouettes were added to the second image. I get a sense that it's trying to mimic something, but I don't see how it relates to the prompt.
Selected images
Here are a few that stood out for me:
Prompt 1 // Image 1
Why? There's a separation of the canvas into 3 zones. The top burst and swirl of paint, the middle "impact" zone which is darker and more horizontal, and then the relatively clean bottom with the "drips". I quite like the image, and it was the first image generated by the prompt.
Prompt 1 // Image 12
Why? This was one of the first to vary from the others. It's almost a landscape on fire, with smoke and reflections in a foreground lake. It's probably drawing from something similar in the training set.
Prompt 3 // Image 3
Why? This one stood out also due to its lack of similarity with the others. It's more similar to clouds than the others (or, billowing smoke). I'd be happy to have this on the wall.
Conclusion
Well, the prompt was minimal and abstract, and I shouldn't be surprised that virtually all the images were incredibly similar.
But as a first experiment with generative AI tools, I'm happy with the results. Which also means... I can understand why people are both excited and scared about the tech.
Ethics?
There have been lots of debates around the ethics of this. For example, if an AI model was trained based on the work of real artists, is the resulting work somehow derivative? Should the original artists be compensated in some work? Is the new work truly original or is it somehow dependent on the original artists' work?
I can't say I've thought about it enough to have any strong or nuanced opinion. Let's do more experiments in the upcoming weeks and see where we end up.
// ka
