MLRUG

I started this project a couple of years before many of the impressive AI tools (Stable Diffusion, VEO, GPT-5, etc.) we have today existed. The only realistic way for me to conduct this experiment was to train my own algorithm (based on other people’s algorithms, of course).

As with so many projects in the field of machine learning, one of the most important considerations for me was the question of what data I would feed into my generative algorithm. I wanted to choose a design object that I was reasonably familiar with, for which I could obtain a sufficient amount of data, and that could be represented as an image (since 3D generative design is much more difficult). Based on these considerations, I opted for Moroccan carpets. I am very familiar with Moroccan carpets because my father trades with them, which also gives me access to image data of Moroccan carpets. I supplemented the data I was able to obtain from my father with images from the internet, which I collected using a web crawler. In the end, I was able to obtain about 3.000 images of Moroccan carpets. This is not a huge amount in the context of a generative algorithm, but enough to get first reasonably useful results.

I wouldn’t call the results perfect, but for the amount of data and technical knowledge I had at the time, I think they were reasonable. At this point, it’s clear that training an algorithm on a small scale like this isn’t really going to replace a designer’s design process. But it can serve as inspiration. And it definitely did in this project.

For this second iteration of the project I reevaluated which algorithm to use and landed on StyleGAN2-ada, a variant specifically designed for training with fewer images. Using my father’s high-quality photo archive again and this time supplemented by much more carefully selected internet images, we achieved much better results than in my first iteration. Showing how much machine learning had evolved in such a short time (and how much of an effect good quality data can have).

The most exciting part of this collaboration with my sister and father was that it brought me full circle, returning me to my original curiosity about using machine learning as a tool in the design process by actually bringing these digital creations back into the physical world.

Rather than choosing Indian or Nepalese manufactories with their precision technology that would allow nearly photographic reproduction, we deliberately opted for production in Morocco. The rhizomatic producer landscape there consisting of many small, interconnected local businesses allowed us to maintain the character and vitality of the original designs while ensuring value creation remained in the country that inspired our work.

Working with a producer from the western Middle Atlas that my father had collaborated with for several years, we were able to translate the algorithmically generated designs into handcrafted rugs. The result was not a mechanical reproduction but rather a living interpretation that preserved the improvisational qualities that make Moroccan rugs special.

These algorithmically generated, curated and handcrafted rugs were presented as part of Design Month Graz 2025.

The project gained recognition beyond the exhibition, being featured in two articles in Cover, a leading publication in handmade carpets and textiles for interiors.

This coverage positioned MLRug within important industry conversations about the intersection of traditional craft and emerging technology. Lucy Upward’s feature article^[1] explored the project’s conceptual framework, examining how AI’s pattern-learning process mirrors the historical migration of motifs along ancient trade routes. Meanwhile, Denna Jones’s analysis^[2] used MLRug as an exemplary case study for ethical AI implementation in design, by keeping production in Morocco and creating a purpose-built dataset to avoid copyright concerns.