LOREM
words by Matthias Schäfer
I first encountered Lorem through the physical artifact of Adversarial Feelings: a book with a cover of a computer generated face with translucent skin; inside a USB stick with tracks by Lorem and music videos created by experimental AI artists. The graphics are otherworldy and weird, the music itself was striking: tracks with all-caps titles, interesting textures, and genre-bending rhythms that felt both precise and wobbly.
When I later discovered that a Lorem model was trained on Neutone Morpho, I immediately reached out to ask if they would be open to a public release. I am happy to announce that LOREM is now available as a model in Neutone Morpho. Trained on an archive of Lorem’s sounds and finished tracks; it pairs beautifully with drums, adding a rumbling low-end but can also transform guitar drones into alien soundscapes.
To celebrate the release, we sat down to discuss what’s behind the Lorem project. I was absolutely astonished by the multiplicity of their output, which ranges from music releases, to opera shows and large-scale art installations. It is a body of work that seamlessly threads human emotion with cold computation.
Can you explain where you come from, what your background is, and how you ended up working with neural networks and audio?
I started experimenting with neural networks and audio around 2016. I have a background in music from an instrumental band (guitar, synth and percussions). After almost 10 years of touring around Europe, we took a break to explore separately. At that time, I was really into sampling techniques, granular synthesis, and stuttering music, using different samplers, including the Orthogonal Devices ER-301 module. I was visually and conceptually intrigued by archive-based practice, constantly re-sampling my own material to transform it into something new. I was also experimenting with MIDI “sampling,” which is how I stumbled onto an old-school LSTM model on GitHub. It was designed to reproduce Metallica’s MIDI drums, so I replaced the original dataset with my own recordings. The idea was to train the machine to play my sampler, generating new MIDI in real-time. My friend Luca Pagan developed a multi-channel sampling software called LAIN in Max, which gave me an opportunity to merge my digital archives with an advanced hybridation tool. From then, I started experimenting quite a lot with ML in music. That’s also how I stumbled into Neutone after meeting Portrait XO, who spoke about her collaboration with you at the Sandberg Institute.
Your first release in this space is Adversarial Feelings (2019). Can you talk about why you made it, how, and why these different styles of publications (Music Album + AV Show + Book)?
When I started the Lorem Project, I wanted a place to let my interests converge, as I had been running a publishing house (Krisis Publishing) since 2010. Beside my musical practice, I have a background in design and visual arts. Machine learning was a useful topic to inspect because it approaches different media with the same perspective and discloses big theoretical and political issues. I tried to tackle the same question across text, visuals, and music: we are dealing with statistic-based technology. Usually, this is used to map and extract information from our emotional behavior—mainly for marketing or security—to quantify behavior and extract value. I wanted to reverse this idea: rather than extracting value, find ways to “perform human emotions” and observe our behavior through the warping lens of a statistical system. It’s an occasion to observe the foundation of our emotions through this weird lens. That remains part of my interest in experimenting with statistics and machine learning.
When you worked on this in 2018, did you already anticipate how AI would control human emotion and shape information with things like large language models in 2026?
No, I was really blind somehow. Of course, some emerging problems were easy to foresee and have now become huge, but I didn’t expect such a quick evolution in the industry and its impact on societies. I didn’t use large language models for Adversarial Feelings; it was mainly recurrent LSTMs, RNNs and primitive technologies. They were very creative because they were glitchy and deeply related to the datasets I used. The error rate was high, but that was charming, especially since I was into random patterns already. It embedded randomness in a slightly more intelligent way. ChatGPT’s release in 2022 changed everything—I was totally unprepared for that. I initially thought the slow process of training would mean my practice would grow slowly over time. Instead, it happened very fast. The way we approach models now, specifically regarding user interfaces, was science fiction back then. I didn’t expect that my mother would soon be interacting with the same models I was experimenting with.
You also worked a lot with Generative Adversarial Networks (GANs) to create visuals for your work, right?
I still use those small, old-school models with students because the human-machine interaction is more vertically connected with the datasets and your curated archives. This connection is missing now; modern interfaces are increasingly detached from the actual data. That’s also why I’m interested in Neutone—you offer accessible tools for musicians without deep ML knowledge, while still enhancing the role of data curation. You keep the relationship with the machine transparent, which is often missing in what they call “generative” tools, especially in music. Text-to-audio systems where you have no clue where the sound comes from are pretty problematic, I think.
Other artists have told me they would never release their samples, they view their archive as an instrument they curated over the years. What is your take on archives and authorship in this field?
I totally understand that. For me, the archive is difficult to separate from the artwork; it is simultaneously the instrument and the artwork. Working with this technology raises big questions about authorship: is my archive really mine, or is my authorship just the convergence of objects I decided to collect? However, I’m happy for people to experiment with my sound if I release a model. I’m not scared that people will play music that sounds like mine. I distrust the idea of authorship as something that comes from an artist’s genius producing a completely new idea from scratch. I believe artists are creative when they connect unexpected points, and through that, they can say something new. It’s interesting if my practice becomes part of a wider narrative within other’s work. That’s what being “intertextual” is somehow: letting people take your instrument and use it in the “wrong” way or in new directions.
Can we talk about how you work with collaborators? You seem to reach out to many other artists that are interested in both visuals and music.
The first collaboration was triggered because I had no computer science skills. Finding resources for artists wanting to approach data science was hard. A crucial collaboration was with my friend Nicola Cattabani, a physics PhD who switched from nanotechnology to data science. As he trained, I constantly asked for his help, which helped me understand. It was interesting working with someone highly skilled yet still learning—it made the technology more accessible. After a couple of years, I became more autonomous but still involved him in early projects like Adversarial Feelings.
From there, I mostly collaborated with artists not trained in technology but who had specific skills I wanted to embed in my projects. One important collaborator is Karol Sudolski, a digital artist skilled in photogrammetry and 3D rendering; we worked on hybridizing machine learning with visuals and he’s almost a fixed collaborator of Lorem now. Coming from a band background, collaboration is vital to me. Most of my important work with Lorem was based on collaboration, which led me to meet Danny Elfman and work for Blixa from Einstürzende Neubauten, one of my heroes. Interdisciplinarity is an exciting way to both make my work more deep and to connect with human beings.
Right, art is about connecting with people. Your last album was Time Coils (2024), but you’ve also expanded into film and installations. Can you talk about that shift?
Adversarial Feelings was primarily music-driven and intended for the music industry. Then COVID hit, and touring became impossible. I began designing live sets that felt closer to watching a Netflix series than a standard AV set with abstract visuals. This led to invitations from film festivals like Sheffield DocFest, where I realized the potential of filmic and installative versions of my work. My practice expanded significantly into the art field, leading to my first large-scale installation, Distrust Everything (2021). From then, artistic projects became an important component of Lorem’s identity. But I also worked on collaborative releases between the two albums.
Can you explain what the installation is about?
Distrust Everything is a four-channel AV installation with a quadraphonic PA and a mirrored floor for immersion. It’s based on 21 years of an American artist’s dream transcriptions. The models use this dataset to generate an evolving, synthetic dream screenplay, adding elements to create a larger, connected narrative universe.
That sounds amazing. Lastly, what are your current projects and your main research focus?
I am currently focusing on an opera show, my second collaboration with Italian director Silvia Costa, premiering in Amsterdam in March. I am also working on a new band project with techno/experimental producer Katatonic Silencio and hopefully continuing a collaboration with Pinch from Tectonic Record. My main research interest at the moment revolves around the topic of angels, specifically focusing on the historical corpora of information produced across centuries and different cultures regarding the figure of the angel. I am exploring how to relate this historical and theological research to technology.
It was a pleasure to dive deeper into the technical and philosophical aspects of Lorem’s work. And it’s refreshing to see an artist, who comments so clearly on AI and technology, creating their own tools and expanding creativity.
Don’t forget to try the LOREM model yourself and see how it fits into your own creative process.
You can follow the work of Lorem here: