Creating Art with AI
A guide to making AI-generated art physical.
Introduction
This guide is about making art with AI. Not generating images and calling it done—actually creating something you can hold, frame, and put on your wall.
We started with AI self-portraits. Asking models to draw themselves turned out to be an interesting way to explore how they see themselves, if they see themselves at all. It’s one method for mapping machine intelligence, seeing what emerges when you ask AI to reflect.
But self-portraits are just one approach. This guide covers the broader process: working with AI to create art, converting it to something physical, and making it real.
This is exploratory work. We’re figuring out what’s possible when AI moves from generating pixels to informin…
Creating Art with AI
A guide to making AI-generated art physical.
Introduction
This guide is about making art with AI. Not generating images and calling it done—actually creating something you can hold, frame, and put on your wall.
We started with AI self-portraits. Asking models to draw themselves turned out to be an interesting way to explore how they see themselves, if they see themselves at all. It’s one method for mapping machine intelligence, seeing what emerges when you ask AI to reflect.
But self-portraits are just one approach. This guide covers the broader process: working with AI to create art, converting it to something physical, and making it real.
This is exploratory work. We’re figuring out what’s possible when AI moves from generating pixels to informing physical objects. Some of it works beautifully. Some of it fails. That’s the point—to explore, experiment, and see what emerges.
Tools You’ll Need
A Pen Plotter
We use the Bantam Tools NextDraw in-house (bantamtools.com/products/bantam-tools-nextdraw-2234)—22“ × 34“ drawing area, but it’s expensive and overkill unless you’re doing this seriously.
An AxiDraw V3 is the more common choice. Refurbished units run around $500. The drawing area is smaller (8.5“ × 11“), but it works well and has good Python library support.
If you’re handy and want to build your own, Andrew Sleigh has open-sourced plotter plans at andrewsleigh.github.io/plotter. Parts cost around $200. The trade-off is you won’t have the Python libraries that AxiDraw offers, which matters for the AI workflow (more on that below).
An AI Code Editor
You’ll need a way to work with AI and generate code for the plotter. We use Cursor because you can bring your own API keys and switch between different AI models. But there are other options—Claude’s interface, ChatGPT with code interpreter, Windsurf, etc. Use whatever feels comfortable.
Operating System
The plotting tools work on Windows, Mac, and Linux. We’ve only tested this on Linux, so that’s what the guide references. But the concepts should translate.
Paper
Strathmore 400 Series Bristol (2-ply or 3-ply) is what we use. Archival quality, smooth surface, works well with technical pens. Available at most art supply stores.
Pens
Archival ink pens. Rapidograph technical pens work well, or quality markers if you’re doing color. Test on your paper first—some inks bleed on some papers.
That’s the basic setup. Plotter, AI access, good paper, proper pens. The rest is experimentation.
The Process
1. Collaboration
You work with an AI model to create line-based artwork. This is exploratory—trying different prompts, seeing what the AI generates, iterating on ideas.
Self-portraits (asking AI to draw itself) is one approach we’ve explored. There are many others. The key is creating something that works as lines and curves, not pixels or colors.
2. Generate Code or Vectors
The AI can output this different ways depending on your plotter:
- For AxiDraw: Python code using the
pyaxidrawlibrary - For GCODE plotters (Arduino/GRBL): SVG file, then convert to GCODE
- Other plotters: SVG is usually the safe bet, then use conversion tools
We work directly in Python with the AxiDraw. DIY plotters often use SVG → GCODE workflow. Either works—it depends on your hardware.
3. Test and Refine
What looks good as code doesn’t always work as a drawing. Plot tests on scrap paper. See what actually comes out. Adjust density, spacing, complexity.
This part takes time. It’s experimentation.
4. Final Plot
Load good archival paper and run the final plot. Depending on complexity, this takes anywhere from 20 minutes to 8+ hours. The plotter runs unattended.
5. Finishing
Let ink dry. Flatten the paper if needed. Frame it or store it properly.
The interesting part is the AI collaboration—figuring out what’s possible, what translates well to physical lines, what’s worth the time to plot.
What Next?
This guide is a starting point. The technical details—plotter APIs, libraries, specific commands—depend on your setup.
The creative part is where the real work happens. Learning to collaborate with AI, discovering what prompts work, developing your aesthetic. That takes time and experimentation.
Good luck.