Making of Colorland Scribble 3

This is Colorland Scribble 3, prints here.

The main idea for this series is iterative variation. I take the latest Colorland Scribble picture, in this case CS 2 (colorland scribble 2), and try something new to create the next one. 

I described a little bit about how Colorland Scribble 2 came to be in the last post.

The starting point for Colorland 3 was the line art from CS 2. I redrew a couple details, but otherwise it's the same lines. The image below shows the line art as a negative.

I was curious about using a generative model like stable diffusion to do something with a variation in this series. So, I wrote some python code. The code takes a random sample of the above line art and feeds it to a controlnet layer in a stable diffusion model. This causes the model to generate an image that conforms to the line art. I then take that image and paste it onto a blank digital canvas in the same location where the random sample was drawn. Then, I repeated that process many times, each time averaging new samples into the canvas as they were generated. 

Some of the random samples look like this:

At the end of the sampling process the digital canvas was full of overlapping samples, and looked something like this:

Next, I brought this into photoshop and treated it like an underexposed image to bring out the color.

The above is...well, I don't know, a colorful pixellated mess. I thought it would be fun to use this as a base palette for oil painting.

So, I took this layer into my main drawing app (Adobe Fresco), turned it into a pixel layer, and then hand painted the entire thing with an oil brush texture. 

Much time passed, and then Colorland Scribble 3 ended up like this: