Matt DesLauriers on Challenging the Image

‍Acclaimed software artist Matt DesLauriers spoke to Peter Bauman (Monk Antony) about the implications of AI's prevalence. They examine how DesLauriers has begun incorporating AI into his process and how AI challenges our relationship with the image.
About the Author
Matt DesLauriers, Filigree (Detail), 2024. Courtesy of the artist and The Disruptive Gallery

Matt DesLauriers on Challenging the Image

Acclaimed software artist Matt DesLauriers spoke to Peter Bauman (Monk Antony) about the implications of AI's prevalence. They examine how DesLauriers has begun incorporating AI into his process and how AI challenges our relationship with the image. 

Peter Bauman: How does your latest project, Filigree, give us a glimpse of what you’re seeing as an artist now?

Matt DesLauriers:
The big thing for Filigree is that I'm trying to integrate AI into my generative art workflow in a way that I feel like I have control over it. Previously, when I've worked with AI and machine learning models, it felt a bit like a black box—you train it and then you don't have much control over it. You don't really fully understand why it's producing certain inputs, which can sometimes produce interesting results. But it leaves me a little bit frustrated. As somebody who's so used to building my own tools, code and libraries from scratch, all of a sudden having to work with this black box is quite annoying. With Filigree, I thought, “What if I can just use part of it?” And it's also asking, “What is the future of this combination of creative coding and generative art combined with the latest AI models?” We're entering this era where AI is becoming more and more useful and functional, to the point that maybe in a few years a lot of coding will just be done using AI systems to produce the code.

How do I still retain some elements of my own signature, my own style within that? That's where Filigree started and it's been a fun and interesting project to dive into in the sense that I'm exploring a few things that I'm not used to. Working with these AI systems is very different from working with the generative workflows that I'm used to and the Javascript workflows that I've been building over the last several years. It's all of a sudden like I'm starting with a few new tools that are a bit cumbersome, a bit awkward or still not quite how I want them to be.

Peter Bauman: Can you talk more about the role that AI plays in the project? Did you train your own models, for instance?

Matt DesLauriers: It's really a light touch instead of going deeply down this rabbit hole of machine learning and training my own systems.

It was more an examination of how, in a few years, many of us will probably be using AI systems that we did not train in the same ways that a lot of us are using computers and phones and things that we didn't build.

At some point, I was interested in trying to build my own machine learning models. And now the amount of compute required and the number of GPUs required are so great to get something really good that it's far beyond my capabilities and what I would want to expend. Instead of trying to build my own, I’m thinking about what I can do with these systems that are being built. How can I use them as a design tool?

For Filigree, it's only about color and tonality. That's the constraint that I worked on with the AI models: producing these images from them and then using them as inputs into my generative tool, just as pure pixel data for color information. That way, I would still have full freedom of control in terms of the composition and the output. But the color and the mood would be determined by an AI system based on some prompt.

Matt DesLauriers, Filigree (Detail), 2024. Courtesy of the artist and The Disruptive Gallery

Peter Bauman: So the palettes come from AI. What was your motivation to incorporate AI into your workflow, as you’ve described? How does AI enhance this workflow?

Matt DesLauriers: The motivation came from a place of book binding, when I was in a book binding workshop at London Centre for Book Arts. It's this real craft involved in making books: folding the paper, stitching it together, gluing. There's this real technique tradition.

I like the idea of combining something that's very automatic and very machine-driven, like AI systems, with this very traditional and very hand-made process. 

I wanted to build a series of handbound books that, in the inner pages, would all use AI-generative imagery but the actual book itself would become this artistic artifact. In contrast, in many art books, the inside pages would typically be the artwork. But the book itself might be mass-produced in quantities of thousands. I wanted to flip that and make it so the images on the inside were mass produced by this AI system but the book itself was hand-bound.

That was the starting point for thinking about Filigree. Then it grew into something larger in terms of screen prints and other things.

As I went down this path, the main benefit of the AI was getting this range of tone and color that is really hard to achieve with a pure, coded generative system that I would typically be writing.

For example, say I want a landscape of mountains; the AI input from that would be something that would take me probably years to develop a single type of image. And that would be very limited to one type of image with my generative system. Whereas with the AI, I could say, “I want to shift that a little bit to something else—like now fields or the ocean.” I wanted to make my use of AI quite limited to color and tonality because I didn't want to lose the work in this AI output. 

One of the things I wanted to avoid was using the outputs of the AI directly. Quite often, these systems will produce four outputs. This seems to be the trend with these latent-diffusion models. It's pretty nice as inputs to a generative algorithm to use these. You can cut segments of each and recombine them to get this scrambled output image, which meshes and mashes these four different outputs together. So the output you end up with is not in any way like a clean or coherent image but it still has these feelings and elements—your brain recognizes a mountain—even though it's a completely incoherent image.

I don't know if you ever did collage with magazines as a kid, where you'd take a bunch of magazines and cut up things and then put them all together. That's what my generative code does, and that's the coded aspect of it.

Image of Matt DesLauriers's screen printing process for Filigree. Courtesy of the artist and The Disruptive Gallery

Peter Bauman: You've said previously to Bright Moments that “generative art is more a system than an image.” But with this project, it seems like you're exerting greater levels of control over the system itself and focusing more on the image by curating the output. In general, where does the art mainly lie in your work—system or results? Is it something that's consistent? Does it change from project to project?

Matt DesLauriers: I don't really feel like it’s too much in the images. There are some projects where there's a single image involved and then maybe it's more about the image. But even then, Sierra wasn't really about the output image. It was more about the process involved with screen printing.

There's always this element of system and element of process that I want to try and highlight.

With Filigree, it's a bit of bringing in this image again and, like you were saying, putting a little stronger emphasis on it. But at the same time, the image is a bit throwaway. That's what I was loosely getting at with the AI systems: I've chosen these different themes and imagery, like mountains or landscapes, but really I could have chosen anything. It's throwaway in the sense that I could just use a different input into the system, use a different prompt or use a different image.

It’s touching on this idea of how the image as a whole might change in the coming years with AI systems and how it's putting a wrench in how we view images and how we create and understand imagery.

Something that's a bit concerning is that traditionally, we would look at a beautiful image and think, “Wow, that's beautiful. How could one person create that image?”

Now with these AI systems, we're looking at beautiful paintings and instead of saying, “My kid could do that,” we say, “I could do that myself.”

I can just punch in a prompt in an AI system. So we have this really different relationship with the image in general nowadays and that will probably continue. I don't want to say it’s a cheapening of the image but the image becomes less important, in my view. The system is still the important thing. And that's where I wanted to focus a little bit more—on this system of cutting and mashing those together to produce something different. Also the system of book binding or the system of screen printing. These systems are the things that I tend to highlight more in the project, as opposed to getting too much into the detail of what exactly the image represents or what exactly the image looks like.

Matt DesLauriers, Filigree (Detail), 2024. Courtesy of the artist and Hanbury Press

Peter Bauman: You mentioned how book binding and screen printing have served as the emphasis or starting point for certain projects. In this way, your code-based work clearly emphasizes the materiality of the digital. How do you also challenge our relationship with the digital image by showcasing the breadth of its materiality?

Matt DesLauriers: With generative art or any art that's software-based, you get this amazing thing that the output can be realized in so many different ways. You could just use some generative system to produce images and then print those images and make art about the images or the prints. But this process of taking a system that can be reconfigured, changed, evolved or iterated upon with different parameters, outputs or iterations is so flexible. It can go into different kinds of print—screen, Riso print or digital printing. It can also go into projection, 3D printing or laser cutting. There are a lot of ways to take this digital system and turn it into a physical system. I’m more interested in the idea that generative art can be made into many of these different systems rather than only focusing on the final output of one physical representation.

In a lot of past projects like Folio, I've talked about how the output is a system rather than a print, an image or even a visual artwork.

That's the one thing that generative art has over a lot of other fields of art, like painting, where you're creating a single static output. With generative art, you're creating a system that can produce many outputs that can be realized in many different media. 

Peter Bauman: We’ve talked a lot about our changing relationship with the image but how do you go about curating your own work or assessing others’? What aesthetic criteria do you use?

Matt DesLauriers: The way I curate is maybe different on this project than some others. For a lot of generative art projects, I'll just sort of let it run and I'll see something, tweak the code and be like, “Okay, I like this.” Or sometimes the code will produce something I didn't expect and I'll go down that rabbit hole, allowing more of that to come into the system. It becomes this nice ebb and flow. The whole time, I’m thinking about the system, the little parameters and the sort of wires that make up the system. With Filigree though, it was a really different and somewhat jarring experience where I would produce thousands of outputs from a particular set of prompts and then I'd have to sift through the ones I liked and determine whether those were satisfactory to what I wanted to represent with this whole art project. And if they didn't meet what I wanted, I would have to start over with new prompts. It wasn't this very seamless workflow.

It was quite stuttered and frustrating in some ways because the AI would sometimes produce these quite stunning images that I could imagine working really well. But then there'd be one little splotch that was a horrible part of the image so I couldn’t use it anymore. So you're starting over but there's no way you can code that into the AI because it's such a black box.

You're just rolling the dice with AI and hoping it's going to work, which I guess we're doing already as generative artists. But usually, when I'm setting up a system, I've set it up so that when I do roll the dice, I know that it's going to give me good results.

If it's not giving me the results I want, I can tweak it. Whereas with the AI, you're shooting in the dark. That was one shift in how I was curating this project versus others. 

I was getting to the point where I wanted an AI to fix these images for me. So you start to get into these AI-versus-AI systems where you have an AI that's curating the outputs. It becomes this real jumble of AI that, in some ways, is making me yearn for a simpler, purely code-based, purely instructional workflow again. 

As much as I feel like AI systems are hard to ignore and they're coming for a lot of coders—they're coming for everyone, I guess—at the same time, I think they don't feel good yet. They don't feel super seamless and they don't give me the same flow state because they are quite hard to control.

Matt DesLauriers, ALICE, 2021. Courtesy of the artist and owned by Le Random

Peter Bauman: You talked about the curatorial process for your own work but what about when you see other artists’ work? How do you go about analyzing it or appreciating it? Is there a step-by-step process that you go through? 

Matt DesLauriers: It's changed a little bit. There was a period where I was really interested in graphics programming so I'd be really interested in shaders and those specific details. I'd see somebody's work that might be using a beautiful ambient occlusion algorithm that just ends up looking like a blender model. But I know what's involved in the process of coding some sort of path tracer so I would be really excited by that. 

Over time, it's changed a bit and I've become less interested in the pure aesthetics and less interested in some of the graphics programming approaches—trying to make really realistic imagery or trying to use these different simulations of light or whatever it would be. I've gotten more interested in conceptual works or process, which it always comes back to. It’s demonstrating craft and building out an interesting system. If somebody posts an image and breaks down how it's made, that's where I'd start to get really excited.

Even if it's something as simple as lines and squares, I'm no longer so concerned about “Oh, that's just a bunch of lines” or “Oh, that's just a bunch of red squares.” Sometimes there's something really captivating about simplicity. It can look simple but you realize it's a bit of a complex system when you run it. 

That's something that's changed a bit: the aesthetic quality is less important now. There are some artists that are really focused on the aesthetics, like the painterly aspect or making it look exactly like some simulated thing. I'm less interested in that and more interested in the code and the system behind it.

For Filigree, trying to get it to work on the blockchain had aesthetic ramifications related to AI. I wanted it to be small in file size because I didn't want to upload thousands of dollars worth of code. I also wanted the imagery from the AI models to be encoded as files. I didn't want to have to upload full images onto IPFS, partly because of the huge file size and partly because the question around copyright doesn't feel solved in a sense. Maybe it's fine right now but in five or ten years, if there's some concern with one of the AI models that I used, I don't want that to be permanently entombed on IPFS and blockchain. 

My idea was to take the outputs, run them through my code and then produce a compressed file that just represents the marks, the drawing marks into the different colors. That file is quite small because I've been able to create my own encoding format for this specific project. That’s a whole part of the project that doesn't get a lot of love because I haven't talked about it yet. It affected the whole aesthetic quality of the work. Before, it was a much more pixel-based output; there were no little, tiny strokes and tiny marks. That's what creates this textile quality. It was more like a 2D, raster AI image. It was just interesting how, in some ways, these processes really changed the aesthetics of the work.

Peter Bauman: You mentioned how it “always comes back to” process for you. How can visual work entice viewers beyond the surface to look deeper into the system? 

Matt DesLauriers: The visual output is like a little teaser in a way. And if it's something that catches the audience, it doesn't need to be the end all, at least in the context that I like to work in. It's not about the image. One image is just one of thousands or millions of outputs that could come from the system. The system is where the heart of the artwork lies and the heart of my focus. You still need to create something that is catchy and that's a hard thing because it's very subjective. Sometimes you create something that works; sometimes you don't. It is tricky because some artists say the image is not that important and then they'll really struggle to capture audiences. Because at the end of the day, we are these visual creatures and we do tend to judge books by their covers; it's hard to get past that.


Matt DesLauriers is an artist whose practice primarily focuses on code, software and generative processes.

Peter Bauman (Monk Antony) is Le Random's Editor-in-Chief.