Plenty of “AI meets nature” projects reach for a generative image model the moment they need artwork. Fugleramme, a Raspberry Pi build from developer Arne Giacomo, deliberately doesn’t — and that restraint is most of what makes it interesting.
Listening, then drawing from the archive
Fugleramme pairs a Raspberry Pi 5 with a USB microphone and BirdNET-Go, the open-source birdsong classifier built on Cornell Lab of Ornithology’s BirdNET model. As birds sing nearby, the Pi identifies the species in real time, then pulls from a curated library of hand-drawn illustrations sourced from 19th-century natural history books on Wikimedia Commons — automatically assembling a small collage of whichever species were just detected onto an e-paper display.
Giacomo is explicit about the choice not to generate the artwork: the point of the project isn’t AI-drawn birds, it’s matching what’s actually singing outside to art that already existed for that exact purpose long before machine learning did. By his own account, “most of the work has gone into deciding on hardware, features and most of all sourcing art” — the classification is almost the easy part; curating and licensing several hundred illustrations well enough to cover a real backyard’s bird population is the actual project.
Seeing it live
The build is deployed and running in Giacomo’s own garden in Norway, with a live feed showing whatever the frame has most recently detected. The full hardware/software project, including the display assembly logic, is open on GitHub.