About this art

Custom-training tools for Stable Diffusion, (a generative image machine-learning model) can be used to teach the model ‘concepts’. I’ve been exploring a use for these tools that I haven’t seen any other examples of yet, but I imagine will become increasingly popular with artists.

Towards the end of last year, I tried to train the model to copy the artistic styles of some of my drawings. I was surprised how easy it was to get it to more or less copy some essence of my style, but the results were more novel than genuinely interesting. I mean, I’m sure I could have improved my results with more experimentation, as these tools do require lots of trial and error. But as soon as I’d tried training the model on the generated images themselves, I saw that this was something far more interesting and started doing only that.

It feels a lot like something else I’ve spent a lot of time doing in my life, looking at art on aggregator sites like Tumblr. It’s always been very clear to me that this process of spending hours trawling through hundreds of images of art, trying to find the ones that do the things, plays a large role in my own artistic practice. But having it merge so completely with the literal creation part of it is very unexpected.

This search for these artworks feels something like learning a language. You keep finding scribbled, in odd corners of the internet, personal statements written in an obscure and cryptic dialect, put out into the world by people like myself. You learn a bit about these individual dialects by decoding some part of the meaning of their messages, and are able to draw on some essence of them when you craft your own. 

Exploring artistic styles with these new tools feels a lot like that process. And I get a similar, strong satisfaction from the images that I generate. It’s of course a very different thing altogether, but the part of me that deeply resonates with the ways that people express things visually, is very, very happy about the state of things.

I’m looking forward to this catching on. I can’t even imagine how differently other artists will use the exact same tools. 

Where to from here?

A problem for me with making these images is that I’ve made many thousands of them, and many of these I like a lot. The images on this page of only snapshots, like proofs-of-concepts. I keep telling myself that I’m going to pause at some point and work on making the images look more finished (the same set of tools offers many ways to work further on any one image), but I just don’t think this is going to happen. The part of me that just wants to see new art stuff is firmly in control.

To manage this situation, I am looking to explore a more ‘expanded’ use of these tools. This artistic process is an application of data-science. The images that I generate, and the records of the decisions that I make in the production process, are also datapoints. Data-sciencists extract meaningful signal from data and in terms of the art that I make with these tools, this signal is something that I get a ton of satisfaction from.