Skip to main navigation Skip to search Skip to main content

Triadic Triptych

  • Kachi Chan* (Artist)
  • *Corresponding author for this work

Research output: Non-textual formExhibitionpeer-review

Abstract

Triadic Triptych is a generative AI animation that reinterprets the abstract costumes of Oskar Schlemmer’s The Triadic Ballet as a contemporary digital performance. Originally premiered in 1922 at the State Theatre of Württemberg, Stuttgart, Schlemmer’s ballet was a symphonic dance of ‘threeness’ — divided into three sections, performed by three dancers in twelve distinct dances, progressing from the playful to the solemn. In Schlemmer’s conception, costumes dictated movement, transforming dancers into sculptural, ambulant architectures.

Triadic Triptych extends this choreographic logic into the generative AI domain. Trained on a curated dataset of over 100 archival Triadic Ballet images, a custom Low-Rank Adaptation (LoRA) model was developed in Stable Diffusion 1.5. In parallel, a procedural character modelling system was built in Houdini based on Schlemmer’s geometric principles, generating 3D rigged characters composed of geometric primitives. Custom motion capture and AMASS datasets were applied to these rigged models to produce kinematic ‘grey model’ sequences, which were then processed through the AI pipeline. The model’s inherent temporal inconsistencies — its shape-shifting — were leveraged as an artistic feature rather than a technical limitation, producing an ever-evolving digital sculpture that shifts and reshapes while retaining the distinct aesthetic of the original.

The work represents a preliminary, experimental study for the On The Other Earth (OTOE) demonstrator at EPFL, exploring the intersection of procedural geometry and latent diffusion models for choreographic visualisation.
Original languageEnglish
PublisherMuseum für Gestaltung Zürich
Publication statusPublished - 29 Aug 2025

Fingerprint

Dive into the research topics of 'Triadic Triptych'. Together they form a unique fingerprint.

Cite this