Holly Herndon & Mat Dryhurst
Starmirror

31.10.25–18.01.26

“Angels appear as what you fear, devils appear as what you desire.”

Holly Herndon and Mat Dryhurst

KW Institute for Contemporary Art, Berlin, and the Kunstsammlung Nordrhein-Westfalen, Düsseldorf partner to realize a major exhibition and live program with artists and technologists Holly Herndon and Mat Dryhurst.

Starmirror proposes a scenario in which artificial intelligence (AI) models coordinate humans to generate intelligence for collective benefit. Created by Berlin-based artists Holly Herndon and Mat Dryhurst, the work extends their long-standing practice at the intersection of art, music, machine learning, and experimental forms of organization. Their projects often propose new digital infrastructures, approaching protocols themselves as a medium—structures that can rehearse new arrangements between humans and AI.

Herndon & Dryhurst have worked with voices and machines for two decades. Since 2016, they have pioneered work with AI choirs, such as Herndon’s album PROTO (2019), which featured “Spawn,” an AI vocal model trained through live community sessions to explore collective singing with machine learning. Starmirror continues and expands this trajectory, transforming KW Institute for Contemporary Art’s exhibition hall into a training ground for collaborative production between humans and AI.

Holly Herndon & Mat Dryhurst, 2025; photo: Frank Sperling

Holly Herndon & Mat Dryhurst mit sub, 2025; photo: Frank Sperling.

In collaboration with the design and architecture studio sub, the artists have created an immersive sound installation that draws on recordings from earlier collective singing projects while simultaneously providing a live environment for new work at KW. At its center is a new songbook generated by a model trained on Ordo Virtutum (1151)—a morality play in which a soul must choose between good and evil. The play is by Hildegard von Bingen (1098–1179), the German Benedictine abbess and polymath. As the artists explain, “We took inspiration from Hildegard von Bingen, who channeled visions from the divine to propose a new harmony between humans and the cosmos. We see analogies between her celestial hierarchy and the stack of influential protocols that determine our culture.”

On several Sundays throughout the exhibition, participants are invited to join a live process of AI training as the KW hall transforms into a recording studio where choirs, an ensemble, and visitors contribute their voices in call-and-response sessions. These recordings will form the basis of a public choral dataset which will be used to train a Berlin AI choir to debut at the Kunstsammlung Nordrhein-Westfalen in Düsseldorf.

Holly Herndon & Mat Dryhurst, 2025; photo: Frank Sperling. 

Before entering the hall, visitors encounter Arboretum, centered on Public Diffusion, a foundation image model created by the artists trained entirely on public domain data. If Hildegard von Bingen was concerned with angelic orders, Herndon and Dryhurst are concerned with the hierarchies of technical protocols and their invisible role in shaping the world around us. Together, these elements form a public AI protocol, where human and nonhuman agents contribute to a shared context. 

Holly Herndon & Mat Dryhurst, 2025; photo: Frank Sperling. 

The exhibition addresses a critical gap in the public perception of AI, offering a direct, participatory experience with the human contributions that define it. AI is us, in aggregate, with emergent properties. The artists challenge the fear that an AI-integrated future will reduce us to passive consumers of endless automation, proposing “agentic social mediation not social media.”

The exhibition design is conceived by the architectural office sub.

The exhibition is a collaboration between KW Institute for Contemporary Art, Berlin and the Kunstsammlung Nordrhein-Westfalen, Düsseldorf, where the second part of the exhibition project will be shown from 27 June – 11 October 2026.

Curators: Emma Enderby, Liberty Adrien
Assistant Curator: Linda Franken

Holly Herndon & Mat Dryhurst, 2025; photo: Frank Sperling. 

Holly Herndon & Mat Dryhurst, 2025; photo: Frank Sperling. 

Holly Herndon & Mat Dryhurst, 2025; photo: Frank Sperling. 

“We are trying to position AI as a monumental collective accomplishment and coordination technology, part of a lineage that goes back to group singing rituals that predate language, and religious protocols that emphasize participation in something greater than the sum of its parts.” – Holly Herndon and Mat Dryhurst

Trailer Holly Herndon & Mat Dryhurst – Starmirror, KW Institute for Contemporary Art, Berlin, 2025. Camera: Michael Breyer, Production & Edit: Studio Schaack.

Related Events

First Room

Arboretum (2025), Ur-Hildegard Training Corpus (2025)

Arboretum centers on Public Diffusion, an image model trained entirely on public domain data. Each motif in the work appears at least 600 times in the dataset—the point at which it becomes a general-purpose concept anyone can use. At its core lies PD40M, the largest public domain image dataset ever assembled, created by the artists and comprising 40 million high-quality images. Herndon & Dryhurst present this work as an alternative to fears of AI turning humans into passive consumers. For them, the problem is not AI itself but the platform economy surrounding it. Starmirror reframes AI as a coordination technology that recognizes patterns in collective human activity and fosters shared intelligence. At the center of Arboretum sits Ur-Hildegard Training Corpus, a songbook based on Hildegard von Bingen’s Ordo Virtutum (1151), its medieval notation reinterpreted through machine-learning models developed with Algomus. The songbook sits in front of a large-scale print of a landscape—one of the most featured motifs on the internet—generated by Public Diffusion.

Hall

The artists created an immersive sound and light installation in collaboration with the design and architecture studio sub that both presents recordings from earlier collective singing projects and serves as a live environment for new work.

The central ladder in KW’s hall weaves together references to religious iconography, architecture, and computational models. Echoing Hildegard von Bingen’s visions of celestial hierarchies and the computer scientist Geoffrey Hinton’s “ladders of abstraction” in machine learning, it links spiritual ascent with the layered logic of neural networks. The installation also includes The Hearth (2024), a musical instrument, an acoustic organ powered by GPU fans normally used to cool the graphics processing units that enable the rapid, large-scale computations behind AI models.

The sound installation draws on recordings and AI-generated outputs from Herndon & Dryhurst’s collective singing projects—ranging from music played through the organ and a collaboration with musician Maria Arnal, to recordings at KW of the Ur-Hildegard model trained on Ordo Virtutum (1151), as well as Linked Diffusion, an AI system trained on the artists’ sound archive and recordings of community choirs.

Live Programme

The exhibition invites visitors into a live process of AI training: on selected Sundays, visitors join the ensemble and a choir to contribute their voices in a recorded call-and-response session, based on Hildegard von Bingen’s Ordo Virtutum (1151). These recordings form the basis of a public choral dataset, which trains an AI Berlin choir to debut at the Kunstsammlung Nordrhein-Westfalen, Düsseldorf.

Find more information here in our exhibition booklet.

Holly Herndon & Mat Dryhurst
Starmirror

Exhibition booklet

Holly Herndon & Mat Dryhurst – Starmirror Training Performance in den KW Institute for Contemporary Art, Berlin 2025. Photo: Frank Sperling.

“We’re looking not only to present a new kind of exhibition but also provide infrastructure – an open protocol – for others to use going forward.” – Mat Dryhurst

Plain Language

Here you can find our exhibition text in Plain Language. 

Artist Biographies

Berlin-based artists Holly Herndon (b. 1980, US) and Mat Dryhurst (b. 1984, UK) are known for their pioneering work in music, machine learning, and ‘protocol development’. Their expansive practice has led to precedent-setting projects where the technical systems that underwrite creative output are artworks unto themselves. For Holly+ (2021), they trained an AI clone of Herndon’s voice that anyone could use freely, with a mechanism to share profits from commercial use of her identity. Herndon and Dryhurst’s critically acclaimed musical works including Platform (2015) and PROTO (2019), released through 4AD, have toured major venues like Barbican, London and Volksbühne, Berlin. Their image making practice primarily documents their novel interventions in software.

Holly Herndon and Mat Dryhurst; photo: Diana Pfammatter.

Crossing the Interface (2021) represented the world’s first text-to-video animations, and Classified (2021) provided a novel means to produce portraiture of embeddings within image models. Herndon and Dryhurst most recently exhibited at the 2024 Whitney Biennial, presenting xhairymutantx (2024), an interactive text-to-image model designed to poison Holly’s public image. In 2024 they presented the solo exhibition The Call at Serpentine Gallery in London. 

In 2022, the duo co-founded Spawning, an organisation building a consent layer for AI, including tools for artists such as haveibeentrained.com, Kudurru and Source.Plus. They were awarded the 2022 Ars Electronica STARTS prize for digital art, Austria’s first Digital Human Rights Award in 2024, and the Kairos Prize in 2025. Holly Herndon holds a doctorate in Computer Music from Stanford University and is a fellow at the Berlin Artistic Research Programme 2024/25.  Since 2021, Herndon and Dryhurst have hosted the Interdependence podcast where they share their ongoing conversations with a network of artists and technologists working with music, AI and crypto.

sub is a Berlin-based architecture office committed to designing spaces of optimal relevance and to creating environments for culture to unfold. Their work is rooted in deep research into sociocultural dynamics, semantics, and technology, and it spans a wide range of spatial and temporal scales.

“The model is the artwork. It’s not the sculpture or the painting. It’s the model that can generate infinite artworks, in any kind of medium.” – Holly Herndon  

Herndon Dryhurst Studio
Design & Fabrication: Ben Creek
Design & 3D modeling: Annkathrin Kluss
Audio Engineering: Ian Berman
Graphic Design: Patricia Klein

Ur/Hildegard Model Research and Songbook Generation: Valentin Crogiez & Mathieu Giraud (Algomus, Université de Lille, CNRS)

Musicology Research and Musical Arrangement: Dr. Thomas Fournil

The exhibition design is conceived by the design and architecture studio sub

Lighting Designer: Bianca Peruzzi

Ensemble members: Sara Gouzy, Marina Kerdraon-Dammekens, Andria Prokopa

Choirs: ‘HXOS Chor Berlin, Byrdland, Chinesischer Akademikerchor Berlin, Haneen Frauenchor Berlin, Kammerchor der Humboldt-Universität zu Berlin, Kammerchor Vocantare, Koreanischer Meari Chor Berlin

Legal Advisor: Katharina Garbers-von Boehm

Public Diffusion, 2025
Spawning: Jordan Meyer & Nick Padgett
Graphic Design: Michael Oswell

Starmirror App, 2025
Graphic Design: Patricia Klein
App Development: Tero Parviainen & Samuel Diggins

Supported and commissioned by Okayama Art Summit

Accessibility Information

Show more Show less

Venue

The exhibition is located in the exhibition rooms of KW (Haupthalle und EG). The exhibition rooms are accessible via stairs and an elevator. The elevator is accessible from the counter area on the ground floor.

Time

The exhibition opens promptly at 11 am and closes at 7 pm. It is closed on Tuesdays and every first Thursday in the month it is open until 9 pm.

In the exhibition c. 5 works are presented.

The duration of the exhibition visit can be determined individually. Viewing all the works in their entirety would take approximately 1 hour.

The total duration of all audiovisual works is approximately 40 minutes.
The longest audiovisual works is 40 minutes.

Seating

A limited number of seating (benches) is integrated into the exhibition.
Seating (stools) will be available on request. If you wish seating, please contact our staff on site.

Sensory Stimuli

The exhibition tends to have a high volume.
The exhibition space is bright.
The exhibition space is darkened.
The exhibition is immersive.
The exhibition contains stroboscopic lighting effects.
The exhibition space is spacious.
Earplugs are provided on site on request.

Funded by the Art & AI program of the Kulturstiftung des Bundes (German Federal Cultural Foundation). Kulturstiftung des Bundes (German Federal Cultural Foundation) is funded by the Beauftragter der Bundesregierung für Kultur und Medien (German Federal Commissioner for Culture and the Media).

In collaboration with the Kunstsammlung Nordrhein-Westfalen, Düsseldorf

With thanks to d&b audiotechnik with Soundscape

The development of Choral AI presented at Starmirror builds on The Call (2024–25), an R&D and exhibition partnership between Serpentine Arts Technologies, Holly Herndon, and Mat Dryhurst.

Holly Herndon is a fellow at the Berlin Artistic Research Program 2024/25.

Media partners