Hope & Despair
What does memory sound like?
Hope & Despair connects seven memorial sites and museums across the Danish-German border region, places that have largely worked apart despite sharing elements of the same history of occupation and deportation.
Visitors to these memorials come with different motivations: historical interest, mourning, education, or something more private. Whatever the reason, the visit can be intense, and everyone processes their impressions differently: a line in the guestbook, a message to a friend on the way home, a conversation in class, or nothing said out loud at all.
This experiment, one of several in the programme, uses sound as another medium for visitors to express what they felt and take part in a conversation that usually runs one way. It was tested for the first time with a class of 15- to 17-year-olds at Frøslevlejren, the former internment and transit camp, now a national museum.
YEAR AND LOCATION
2026
Frøslev, Denmark
PARTNERS
Designskolen Kolding
Frøslevlejren Museum
KZ-Gedenkstätte Husum-Schwesing
Interreg Deutschland-Danmark Hope & Despair
DESIGN AREA
Play Design · Museum Experience Design · AI & Sound
ROLES
Francesco Aloisio: design and development, covering the activity format, the interaction logic, and the image-to-sound pipeline.
Concept by Philipp Cordts, developed with the wider Hope & Despair consortium.
The soundscapes from Frøslevlejren, as they appeared on the screen in the room.
Questions
At its centre is one question: can a machine work as an expressive medium rather than a source of answers, and help people express feelings that are hard to put into words?
Around it sit several smaller ones. Can AI help us understand memory? Can it show a group something in their own work they would not have seen otherwise? What does it add that a guestbook or a phone camera doesn't? Where does it fail, and what does that failure show?
And one about how it is built: can this be done responsibly, privileging open-weight models over closed ones owned by large tech companies? And can this provide a blueprint for how AI could be implemented in small museum settings?
How it works
After a guided tour of the site, participants work in small groups inside the museum. Each makes a collage from copies of wartime newspapers and photographs, in response to what they have seen.
The collages are photographed on phones and uploaded to a web page. A vision model gives each image a title, a visual description and a written sound prompt; an audio model turns that prompt into 30 seconds of sound. The prompts are configured to avoid melody and beat, so the results do not turn participants' responses into entertainment. Both models are open-weight, and every result is stored with its collage and description, so every submission can be checked and, if needed, removed.
The soundscapes appear on a shared touch screen as floating bubbles, one per collage. Participants can open them alone, to see a collage and hear its sound, or together, layering several sounds into one composition built as a group. The session ends with a group discussion of the experience.
The class on the guided tour, below the watchtower at Frøslevlejren.
Collage-making inside the museum, using copies of Danish newspapers from the occupation.
The uploader: a finished collage, photographed and ready to be turned into sound.
The full pipeline: the screen in the room on top, and the phone, the models and the public repository that feed it below.