Mesh Paintings process

Maintaining an art studio long-term can be seen as a luxury in regards to space, cost, time, family commitments and availability, especially in London. Many artist communities across London have been lost to gentrification. And following my own eviction from an artist-run studio space in London (now luxury flats) I stopped making physical mesh paintings.

I continued my painting practice, on the kitchen table, however, subsequent closure of the manufacturer of the perforated aluminum mesh material, went out of business, and I couldn't find anything similar anywhere. Secondly, the acrylic paint supplier also went out of business - I used pots of Spectrum paints and it was ideal for its consistency, the way it dried - it was perfect. I tried all the other acrylic paint brands, and nothing worked the same way, so I stopped painting for a few years.

Sometime later, in 2020, early 2021, I started experimenting with digital tools to replicate my mesh painting process. I replicated the perforated mesh in Blender 3d so I could try recreating the physical mesh painting process in Virtual Reality (VR) using this new digital mesh canvas and virtual paint. Unlike the physical process, where a large amount of acrylic paint is pushed into the mesh with a pallet knife, the virtual paint is pushed in one hole at a time, using the VR controler and virtal paint.

The mesh painting templates are imported into blender 3D, where I edit the mesh, compositions, colours, positions, perspectives, shapes, sizes and sometimes movement, until I have a desired outcome, which is then rendered as an image, video, GIF animation or exported as a glTF model. For each piece, I use keyframes and the timeline to construct and deconstruct it, and sometimes there are unexpected results, happy accidents during the process.

The move into AI (artificial intelligence) as an art medium felt very much like a natural continuum of my artistic journey. I've experimented with various AI and generative adversarial network (GAN) tools since 2021. My approach has been low tech, fueled by curiosity.

I'm not in any way a technical genius, I like to find and use tools which are easy to use and accessible, which I think is related to my work as an art school educator.

Adopting new and emerging technologies can sometimes be a challenge. I try to grasp a basic understanding and awareness of specific technologies which may benefit my art practice and can do what I want it to do. I gravitate to tech that is accessible, affordable and easy to use. I often drift from one digital tool to the other, without becoming a specialist or master of any specific technology.

For my Artificial Mesh Paintings, I'm interested in exploring ways AI technology can align with my creative conceptual and aesthetic expectations, as well as enriching the overall connection to my art making practice.

Since 2021, I've been experimenting with tools such as Google Colab VQGAN, which gave me some interesting results at the time, however the process was a little overly technical, unstable and fairly unpredictable. Since 2021, I've been experimenting with tools such as Google Colab VQGAN, which gave me some interesting results at the time, however the process was a little overly technical, unstable and fairly unpredictable.

The AI process at this stage, felt more machine driven than human and was not representative enough of my art practice, so I was keen to find a more suitable balance between the handmade and what is created, imagined by the machine/AI.

I had some success in finding the right balance between my art practice and AI with image to image style transfer tools, this is where a reference image style can be applied to another image.

A year or so later I explored purely prompt based AI image outputs, this is where you write a series of text prompts, commands for the AI to respond to create an image. See AI Colour Studies below.

The text prompts reference a huge deep learning, text-to-image model called Stable Diffusion, which is trained on 'pairs of images and captions, publicly available dataset derived from Common Crawl data scraped from the web, where 5 billion image-text pairs were classified based on language and filtered into separate datasets by resolution'.

Exploring AI text prompts was a useful process that made me more aware of the structure and workings of AI. From this point I was more interested in what datasets were being referenced to make these images and their origin.

In September 2023 I came across Runway AI, a tool which provides an easy way to experiment with creating bespoke training models and data sets. I was keen to try and create my own bespoke dataset, so I gathered together all my digital artworks from over the three years, including 100s of mesh paintings (Created in VR and Blender 3D), some minted but many un-minted as experiments or not quite ready for sharing. Once I had the images in a single folder, it was easy to train my own bespoke model and dataset using Runway ai.

With the new bespoke data set I had the freedom to explore new abstract and figurative compositions and subject matter, way beyond the constraints of my digital mesh painting process.

The artificial mesh paintings and artificial landscapes are created using various text prompts, which reference the bespoke mesh painting data set together with the Stable Diffusion model, depending on the specific strength of the parameters.

The Artificial Landscapes move between abstraction and figuration, the text prompts provide a familiar context which is juxtaposed against the unusual abstract aesthetic referenced from the mesh painting dataset. I like to explore the relationships and limits of blending my bespoke dataset with the public datasets, the process feels similar to making a collage. A process which takes me right back to the beginning of the mesh painting process when I created collaged landscapes by assembling paper strips from billboard posters, filling in the gaps with acrylic paint with mesh. The AI often montages segments of photographic content with the abstract shapes, patterns and textures of the mesh painting outputs.

New code-based mesh paintings are created using p5.js and processing. I’m not a coder, but I learnt the basics of code many years ago, mostly for web development, so I understand some of the principles of code. However, generative art code development has always been out of reach for me, beyond my skill set.

So, as part of my AI experimentation I started to explore the use of AI to support the creative coding process, which was a huge game changer for me. Exploring these new generative mesh paintings is a natural next step/iteration of my arts practice. It's taken over 2 decades of art and mesh painting experimentation to get to this point.

My first Pen Plotter Mesh Painting, made with Processing, saved as SVG. Test printing 2 layers, 2 colours, A1 scale and positioning, using @bantamtools NextDraw 2234, large format pen plotter with Promarkers