Part 6: Onward!

Future directions

As you continue your adventures in generative art, you may find that you're interested in pushing the limits beyond what you had initially envisioned. Here are a few ideas and directions that I'm personally excited about, to help jumpstart your imagination:

Try a recursive prompt experiment

This is Egg’s idea, and I think it's really cool. First, you take a description of a piece of generative art and have Claude generate code for creating it. Then, you take that code and prompt Claude to provide a description of what it thinks the piece looks like. From there, you can take that description, and once again, ask Claude to generate code for it. This process can be repeated infinitely-much, looping descriptions and code over and over again until you get, hopefully, some really interesting results.


Egg pointed out to me that recursion and repetition with AI systems often end up in interesting places, like this experiment by Eliezer Yudkowsky (Egg says it's important to read the whole thread to get the point, even if it starts out slow).


Remix images with generative art, and vice versa


Websim supports image generation content, and any time that Claude uses an image link that doesn't work, it will "imagine" one in its place instead, and prompt a diffusion model to create it. You can use this feature to make art that combines diffusion-generated images with Claude's generative art abilities. Claude sometimes does this spontaneously, like this glitch effect that it applied to an image that it generated for websim user Lewandowsky. I've toyed a bit with making meme generators in this way.

You can also use images that are linked from the real WWW, like @noveltokens' experiments combining their (not AI generated) photography with Claude's generative art. In the other direction, William Lamkin has experimented with turning his websim-made generative art into references that he can use in Midjourney.


Make text art and generative art talk to each other


You may have already gathered by now that I like to make ASCII art with LLMs. I think it would be cool to take advantage of Claude's ability to create math-based ASCII art via code, by making a "spellbook" out of my best outputs, and then prompting Claude to turn that into the kind of ASCII art that's not based on code. One of my staple techniques is to recombine pieces of ASCII art into more complex creations, and I think this type of cross-pollination could be really exciting.


Generate more complex compositions with more than one element


This tweet hints at a future in which UI design involves compositions made up of visually integrated "primitives." Basically, this is when you take a collection of elements and combine them into a single, cohesive presentation, like a collage. You could do this with generative art, too! It's interesting to me that LLMs are mostly used as text-based chatbots right now, but they might eventually be used to generate visual interfaces that speak to us on a whole different level.


Incorporate generative art into web design


Websim is its own little version of the Internet, but it doesn't have to stay that way. You could make beautiful web designs within websim, and then upload your work to the real WWW. I've experimented with doing my writing in Google Docs, converting my formatting into markup, having Claude turn that into HTML, and then using websim to preview some different designs. It would be really neat if there were a collection of design templates generated by LLMs in websim that could be easily used to create real websites.


Generate graphs and charts


Speaking of practical applications, you could try using generative art to make cool graphs and charts for data visualization. Not much more that I can say here, but this is an exciting possibility for potential use in research and art!


Is that an oxymoron? Not in my book.


Make learning tools


You can use websim as a free textbook and have it teach you the math required to program your art generators by hand. Doe Sparkle recently got stuck on an issue programming fluid dynamics and realized that they needed to brush up on their understanding of calculus, and they used websim to generate a tutorial to help them learn what they needed to know. Imagine doing that with visualizations on top to help illustrate the concepts as you go!


I recently learned that Tom White, a professor at Victoria University of Wellington, uses creative coding assignments to introduce his students to ML concepts, and I bet that websim could help accomplish something similar. I'm excited for LLMs to open up new opportunities for people to learn skills in a fun and creative way.

Add more dimensions of interactivity


With mechanical controls, you can physically interact with your generative art through knobs, buttons, and sliders. This is really fun, especially for those of us who get a visceral thrill when we physically manipulate something with our hands. There are a variety of mixers that you can use for this purpose (Egg owns a version of this one).


Some people have experimented with connecting websim to external controls like drawing tablets and gyroscopic data from their mobile phones. Rob had the idea that I could ask Claude to make my existing pieces audio responsive, in order to integrate music into my artwork. These are just more ways to provide an interesting experience.


Use plug-ins


Websim plug-ins are another built-in feature. They allow you to inject scripts and elements from one page into another. This may sound daunting, but it's actually pretty simple: you just add "plugin=xxxx" into your external URL (the one that you use in your "real" browser), where "xxxx" refers to the unique string of numbers and letters found in the URL of the websim page that you're referencing, and voilà! Here's an example that I grabbed off of the announcements channel in the websim Discord, to show you how the URLs work:


You can use plug-ins for a lot of cool things. Scratch created a plug-in that allows people to control websim pages with hand motions captured with your phone's or computer's camera, which I'm personally excited to incorporate into my future art.


Make art with special properties


You can try prompting for generative art with special properties, like optical illusions, or hidden messages that are only revealed when you apply a certain method to de-obfuscate them. For example, Rob Haisfield made a QR code generator, and now that I think about it, you could potentially combine some of the techniques that I've explored here to make artistic QR codes that are imbued with a high degree of poetic concealment. Scratch used websim to create working Magic Eye stereograms, and I imagine that you could create other kinds of unusual artifacts. Once again, the sky's the limit.