Realtime Audio-/MIDI-Reactive Visuals for Corpus Viderum
A high-performance, audio-/MIDI-reactive inference pipeline for Ryan Walsh’s Corpus Viderum. Using Apache TVM to compile a StyleGAN model and push realtime, full HD resolution visuals to three free-hanging LCD panels simultaneously on a single RTX 3090 GPU.
"Corpus Viderum" (Latin: A body seen) presents an A.I. augmented reimagining of the sacred music and art of the High Gothic period.
The work consists of a fully improvised, multi-channel musical performance using sounds produced by A.I. models "trained" on medieval choirs and free-hanging LCD panels, creating a triptych of dynamic stained-glass. The windows display material generated by additional A.I models trained on thousands of photographs of Gothic glass panels. Light pours through these windows and is transformed, flooding the space with ever-evolving colour.
Corpus Viderum is a roughly one-hour long performance of sacred "machine hallucinations". The work both envelopes the audience in a dense vocal soundscape, as well as immersing them in kaleidoscopic light reacting in real-time to the soundtrack's dynamics. The space and the audience are integrated into the work, becoming canvases of light.
More details coming soon…