This post is about the recent emergence of AI artists like Dall-E, Google Imagen, Stable Diffusion et. al. and how to deal with it. My way of dealing with it at least.

^{Stable Diffusion:
A steampunk oil painting in style of Rembrandt of a beagle in a blue hat wearing green Victorian era clothing with a bitcoin emblem having a glass of wine in one and a rose in his other hand in front of his log cabin.}

In a previous post I was pondering about the disappearance of commissioned artists due the prevalence and absolute ease of use of computer generated art. Since you can now ask for just about any kind of art to be delivered on a prompt, it pretty much makes the artists themselves obsolete.

Or does it?

Clearly not. At least not yet. If you check the image above, you will find many mistakes the AI made, as the picture is not consistent in all the details it was asked to perform. Also, if you count the right hand fingers, there is still much to learn. For the AI I mean.

And even if it eventually gets perfect...

Lizzie, a Java based Leela Zero frontend

I am a go player, and in the previous years the game of go and especially the professional players experienced a similar thing when AlphaGo beat each and every one of them in an even game. And I guess computers are even stronger now.

Guess what. Professional players didn't go extinct. They are still playing games, and teaching go. But now they are learning from the computers. They are playing against AlphaGo, Katago, Leela Zero and other neural networked go bots. Not only that, they are using them to go through their old games and as means of teaching new things to their own go students.

I gather that while it is now essential to learn from the best (like computers), it is always nicer to play and interact with humans instead. That's basically why many players still prefer to play against real human opponents instead of computers. (Also, it isn't fun to get beaten up in each and every game.)

I also guess there is still demand for real organic paintings and art. Maybe in the form of watching the artist at their work, and then receiving a physical copy. Or something to that extent. The human element is still something I value over generative "art", and I guess you would too.

^{DreamStudio:
A steampunk oil painting in style of Rembrandt of a beagle in a blue hat wearing green victorian era clothing with a bitcoin emblem having a glass of wine in front of his log cabin.}

So I came up with a theory...

What if...
I could use a neural network to teach myself to draw and paint, digitally at least?

I must admit I'm not very good at drawing or painting. I know the basics well as I've studied arts for years, but I haven't had much general experience to get really adept or proficient in them. Basically my motor skills aren't that good.

My idea to test out was the following:

1. Draw a rough sketch. (Doesn't matter if it's awful.)
2. Upload it to DreamStudio
3. Tune the settings (minimal resolution, multiple results) and describe it.
4. Download the results and choose the best one to improve upon in either paper or in Krita.
5. Rinse and repeat until the picture is ready and/or I get good.

I will use the results the computer gives pretty much as an advice, and try to re-learn some of the basics myself, as the neural networks clearly lack in perspectives and anatomy (both in human and animals). Now there's a glaring hole in the capabilities of AI artists.

Anyhow, I'm not expecting much, but I think it'll be a fun journey anyhow.

I'm now at the third iteration, and here is where I started from. I drew a rough sketch with Krita. (Drawn on a Wacom drawing pad I've practically never used, and it is absolutely horrifying thing to draw on!):

^{A sketch of a rabbit in front of a car and a log cabin holding a glass of wine.}

Here are the pictures DreamStudio came up with based on my original drawing:

I had forgotten the wine glass from the prompt, so I tried again.

I liked the addition of the grass and two little triangles in one of the pictures, so I chose that one for colouring. I also added one more triangle, and decided they can represent mountains in the distance. Granted, there were many log cabins that looked a lot better than the one I chose. But for the sake of progress I didn't really care that much yet.

Here's the picture after I coloured it. I know I could have started adding colour and features to the ground, but I guess that can wait for tomorrow.

Here is the batch of the next iteration. I'll have to choose one of them to work on tomorrow.

I might end up having to cut out the best parts of each and use them as my base picture for the next colouring and improvements.

But I think I'll be off to bed for now.

See you guys later!