We met Mattia Carretti, co-founder of fuse* (with Luca Camellini) and Matteo Salsi, Designer and Creative Coder for this multidisciplinary collective’s large team in the backstage of Videocittà, the Roman festival conceived by Rutelli dedicated to digital culture and hosted for several years now in the evocative industrial archaeology spaces of Rome’s Gazometro. Together, we discussed about Onirica (), a project launched in 2023 that on the day we met, it was about to be staged at the Videocittà.
In all versions of this series, Onirica () has literally entered the dimension of dreams through the use of algorithms that return data relating to sleep in the most diverse forms, in this case thanks to the collaboration with two important dream banks: that of the University of Bologna and that of the University of California Santa Cruz.
After the installation versions created for the INOTA Festival in Hungary and the Alberto Peruzzo Foundation in Padua, Onirica () arrived at Videocittà as a live performance, in collaboration with Diego Trotelli, choreographer of the Aterballetto National Choreographic Centre.
Mattia Carretti and Matteo Salsi shared some peculiar aspects of the backstage of this project, from the selection of data to its formalization in various forms. In recounting some crucial choices, they also delved into the heart of their design methodology as a multidisciplinary team.
Elena Giulia Rossi: I’d like to start by asking about Onirica (), the project you developed for Videocittà, as well as the entire series in progress created to explore the dream dimension and its encounter with algorithmic logic.
Mattia Carretti (fuse* co-founder): Onirica () is a project that emerged from the idea of bringing together the world of AI, through machine learning techniques, and a purely human experience: dreaming. We actually began working on the project in 2022, though it stemmed from research that started much earlier, back in 2018, when we were experimenting with a rather rudimentary AI system. Since then, of course, the expressive possibilities have evolved exponentially alongside technological progress.
At first, our exploration of some data analysis techniques and their translation into images was part of a research process parallel to this because we weren’t yet aware of two databases we eventually worked with: one from the University of Bologna and the other from the University of California Santa Cruz. When we discovered them, we thought that this material on dreams organized in a dataset was ideally suited to connect dream reality and artificial existence.
We initially began the design process with Matteo on the visual side and then brought in a broader team, as we usually do in our collaborative work, combining different skills. Several ideas and possibilities emerged. Thanks to the INOTA Festival in Hungary first, and later to the Alberto Peruzzo Foundation (Padua), we had the great opportunity to have a space in which to experiment.
Having access to space allowed us to choose what worked best in context and this helped us to advance in our research. The first version took form of an installation, created both for the Alberto Peruzzo Foundation and for the INOTA Festival.
This version used technologies available at the time, which we felt were right for the kind of narrative we wanted to tell. The sequences of images transformed into each other, creating a sense of hallucination and stream of consciousness very similar to the experience of dreaming.
Later on, thanks to technological progress and the ability to generate images in real time, we thought it would be interesting to do something live – an idea we had considered from the beginning, that of working with a dancer. What followed was a chain of events and opportunities. That’s how we met Diego Tortelli, choreographer of the Aterballetto National Choreographic Centre, with whom we had long wanted to collaborate.
How did you engage with these databases? In what ways, and where, did their archival criteria intersect with yours?
Matteo Salsi (Designer and Creative Coder): The very first thing we did was to try to understand, on our own, what this dataset was. We’re talking about over 28,000 dreams. It was impossible to read them all. Today, technology allows us to synthesize this kind of information visually. We can organize these dreams in a three-dimensional space, arranged by meaning – and that’s literally what we did: a cloud of points where the dreams were grouped into thematic clusters. This allowed us to identify the most recurring themes in the dataset with greater insight than a random navigation of the data could offer.
We extracted recurring themes based on words that recurred more frequently than others. We wondered how we could convey the idea of something so vast. That’s when the curatorial work came in. We began to connect the dots and trace a narrative thread – not one that encompasses the entire dataset, but one that draws out the story we wanted to tell.
Earlier, you mentioned your working methodology. Can we go back to that? When I visited your studio, I was struck by an incredible energy – the presence of many professionals from different backgrounds, all interested in listening to each other. Can you talk a bit more about this aspect of your design process?
Our teamwork is completely synergistic. Even though there are people who are responsible for individual parts of the project, there’s a constant exchange of ideas between us. We often study a topic, collaborating with scientists too, investing in research, and then come up with as many ideas and experiences as possible.
Like many of our projects, Onirica () is constantly evolving. It’s taken different forms: some more physical for galleries and museums; others more performative, such as the one designed for Videocittà. We will continue to explore other directions in the future. This is a recurring feature of how we work.
At this point, I’d like to explore the merits of the sonic and musical dimension. What role did it play in the overall orchestration of the work, and across the various versions?
In the first version of Onirica (), the installation, we wanted to leave more room for the spoken word.
Working with researchers we were often told that emotionally charging the narration of a dream could, in some way, alter its meaning. Of course, we weren’t completely neutral – we’re artists, after all. But we also want to strike a balance with the scientific aspect. In addition, in this first version, the work we did on the voices was very interesting. We wanted each dream to be voiced differently, to reflect the person dreaming: a voice of a child, a woman, a man, an adult or an elderly person.
At first, we considered using real actors. But we soon realized that it was practically impossible to achieve such a variety. Not to mention the very high costs involved in this type of process.
So we used artificial voice-generation systems. During the production phase we were confronted with several challenges: artificial voices also have hallucinations. Out of ten trials, one was good. Some voices changed tone, others would begin to scream or veer off in unexpected directions. In the end, however, we found that, in each successful trial, the performance was very realistic and effective, much better than the performances of real actors.
In the performative version, we reintroduced a very characteristic musical atmosphere that we had excluded at first. So, some key narrative ideas, which we had set aside in the first version, were later taken up in subsequent iterations.
What kind of rhythm structures the narrative of Onirica () in its various versions?
Mattia: A full sleep cycle, which includes several phases (non-REM and REM), typically lasts about 90 minutes. So, 90 minutes of sleep represents an almost complete cycle, and over the course of an eight-hour night, we go through around 5 sleep cycles. The first phase of each cycle involves a deep dream, while the last phase is always a REM dream.
We knew during which stage the dreams of the dataset of the University of Bologna took place. Researchers would wake patients at very specific times during the night, when particular brain activity was detected, using EEG and other electrodes attached to the volunteers. These indicators told researchers in which sleep phase the subject was in at a given time. Upon waking, the volunteers described what they had been dreaming at the time of waking up, which was then transcribed verbatim by the researchers.
For the installation version of Onirica (), we literally followed this temporal structure. The work is divided into five cycles, just like during a night of sleep, keeping the deep dreams in the deep sleep phase and the REM dreams in the REM phase, cyclically repeating this pattern. The idea was to represent a journey through dream space and the moment of awakening.
The performance is also divided into five chapters. Each of these moments recounts a different kind of threshold type. For example, the first chapter focuses on dreams where you’re alone; the second on situations where you meet other people – family, friends, etc.
The third chapter deals with nightmares; the fourth with people who are no longer with us; and the fifth explores the sensation of awakening while sleeping. In each of these chapters, the music addresses the emotional aspect. The voices, in this case, were created differently from the installation: they were all recorded by the same person, Matteo Amerena, a member of our team. Even in this version, combining voice and music was not easy to manage. These are rhythms that need to weave seamlessly with everything and. above all, with the choreography.
In Onirica (), Diego Tortelli’s choreography with the dancer Hélias Dorvault plays a central role. How does the body enter into the narrative and the dream?
Mattia Carretti: There’s no doubt that in the live version, the choreography and the presence of the body dramatically increase the overall complexity of the piece. In the performance, the dancer was given complete artistic freedom of movement and expression. Of course, there were choreographic choices shaped by the story we wanted to tell.
The choreography followed the story and was not tied in any way to the installation version. At first, we asked ourselves why we should put a body on stage – what did we want to say with this additional element.
And, in my opinion, the study done with Diego Tortelli was very interesting. We asked him to reflect on two key elements in relation to how the body behaves during sleep. On the one hand, there are nocturnal myoclonus – those involuntary movements of the body that happen when you feel like you’re falling and you wake up with a start. These are minor sleep disturbances. We all suffer from them to varying degrees. On the other hand, we considered how the body is perceived within dreams – with its multiple possibilities of movement that would not otherwise be possible in reality.
We worked choreographically across both of these dimensions. Hélias then brought his own interpretation into these directions.
Music, sound, dance and body are all vital and very important components of this project. So too is the visual side and the prompt design work that contributed to the creation of the images. Can you tell us more about that? What does prompt design mean, and what challenges did you face?
Matteo: Prompt Design was definitely a challenge. The tool itself is very limited. We often found ourselves having to reinterpret the story of the dream. What we received from the datasets was a textual report. The big challenge was translating this text into a visual and sound dimension – and image generation models often don’t produce the kind of visual results we expected. So we had to find strategies to “trick” them.
I remember one prompt where there were these egg-shaped children on the deck of a ship, diving into the water among the fish. The prompt turned into a convolution of eggs in the sea with arms and children dressed as eggs for carnival. The model wasn’t trained to recognize that kind of imagery. So very often, we had to find a way to guide the model where we wanted it.
Rather than using a pure text-to-image generation, we worked with an image-to-image approach. We started from a base image, a kind of Rorschach inkblot, and then asked the model to interpret it. This became the key for using the model throughout the performance.
Mattia: It was truly a challenge to arrive at something that felt right for the project. Most of the time, the images didn’t come out very well. They were very stereotyped.
You realize how much these models suffer from many biases – biases that we ourselves, Western human beings, carry with our culture and ideals. We understood that these images, while at first glance very beautiful, were actually too perfect, too uniform. They lacked soul. They didn’t captivate us as much as they fascinated us when we tried, a little at a time, in many ways to hack the system, to modify it, to create our own pipeline, to work to bring out something more original.
We learned a lot and realized that this technology could be a great tool. That’s why we also felt it was important to share what we’d learned – especially with people who are understandably intimidated by this kind of progress, knowing and evaluating its risks and opportunities. That is what we tried to do at the Peruzzo Foundation. Alongside the installation, we also tried to explain what we learned throughout the process.
How did you engaged with the space and context of Videocittà?
Matteo: We have a very close connection to this place because in 2022 we created the Luna Somnium installation inside the Gazometro. It was the first installation ever realized in that space. It was an extremely important experience for us. Beyond the project itself, we understood the impact that art can have on a community. Giving the place back to the community through an artistic project was very important and, for us, very exciting.
So now we have a bond with this space that’s also quite personal. Having the opportunity to return with another project, again near the Gazometro, was very exciting. With Onirica (), though, we were faced with a very different context from what the project was originally designed for, which was a theatre or an enclosed space with controlled lighting and sound. Here, we knew we’d be outdoors where some elements are beyond our control.
In the end, the performance went extremely well, and we were very pleased with the result – despite the fact that, as mentioned, it was a different location and context from what we had imagined for Onirica (). This performance is a work that, as you’ve probably realized, has a very specific structure and narrative, with distinct rhythms and climaxes, so it can feel a little strange to have an audience that comes and leaves freely. At the same time, despite that, we have received extremely positive and very interesting feedback. The more we see the show on tour, the more we relate to the audience and the more we realize that it is a work that also has several unusual, almost unsettling moments. Just like in the dream world: this succession of visions and mental images cannot be controlled and, sometimes, we find ourselves experiencing moments of extreme sadness or fear, happiness or absurdity – it’s like being on a roller coaster. That’s exactly what we wanted viewers to experience, and we’re happy that the performance was so well received.
Obviously, we want to thank the entire festival team: Francesco Dobrovich, Michele Lotti, and all the wonderful technical crew we worked with.
immagini (all): Fuse*, «Onirica ()», live at Videocittà, 2025
Onirica () is an artwork by fuse* | Direction: Mattia Carretti | Executive Production: Mattia Carretti, Luca Camellini | Choreography: Diego Tortelli | Performer: Hélias Salvador Dorvault | Music & Sound Design: Riccardo Bazzoni | Head of Visual Design: Matteo William Salsi | Software Development: Matteo William Salsi, Alessandro Mintrone, Matteo Amerena | Dream Selection: Alessandro Mintrone | Dream Voices: Matteo Amerena | Prompt Design: Alessandro Mintrone, Matteo William Salsi, Mattia Carretti, Matteo Amerena | Light Design: Stefano Cane, Matteo Amerena | Technical Direction: Matteo Amerena | Hardware Engineering: Matteo Amerena, Matteo William Salsi, Alessandro Mintrone | Communication and Copywriting: Virginia Bianchi | Production Assistants: Filippo Aldovini, Virginia Bianchi, Martina Reggiani | Photo and Video Documentation: Matteo Torsani, Emmanuele Coltellacci. The visuals of Onirica () are based on a pipeline integrating the Diffusers: state-of-the-art diffusion model library developed by Huggingface and OpenGL Shading Language (GLL). The performance is powered by NOITOM’s MOCAP Perception Neuron.
Founded in 2007, fuse* is an artistic studio dedicated to exploring the creative potential of contemporary technologies and their profound influence on how we live, think, and connect with the world. Led by founders Mattia Carretti (b. 1981, Italy) and Luca Camellini (b. 1981, Italy), the studio brings together a multidisciplinary group of artists, architects, engineers, and designers who collaborate to create innovative projects, artworks, shows, and exhibitions while simultaneously dedicating themselves to research and experimentation. Inspired by the observation of social and natural phenomena and fueled by scientific discoveries and research, fuse* works evolve through collaborations with specialists and research centers, integrating advanced knowledge and diverse ways of thinking.
Known for large-scale installations and live performances, the studio constantly experiments with new relationships between the physical, digital, natural, and artificial, exploring a wide range of artistic media, including sculpture, print, video, light, and sound. Visit fuse*’s website to know more about the studio and its team.
