top of page

Expanding Multimedia to Smell: Manufacturable Text-to-Scent Generation for Multisensory Holodecks

Jooyoung Kim

Generative AI has made it increasingly common to translate text into images, music, speech, and video, yet smell remains largely absent from today’s generative multimedia systems. Unlike pixels or audio waveforms, scent generation requires a physically manufacturable specification: a recipe composed from a finite palette of ingredients, structured into top, middle, and base notes, with constrained mixture weights. In this talk, I will introduce our recent work on manufacturable text-to-scent generation, which combines large language models for semantic interpretation with a rule-based formulation engine that enforces physical and compositional constraints. I will also briefly discuss ongoing perceptual evaluation results based on 100 manufactured scents and 1,000 human evaluations, which suggest that generated scents can carry weak but measurable traces of their originating textual meanings. Finally, I will outline how manufacturable text-to-scent generation can become a building block for future multisensory holodecks and olfactory interfaces.

Abstract

Generative AI has made it increasingly common to translate text into images, music, speech, and video, yet smell remains largely absent from today’s generative multimedia systems. Unlike pixels or audio waveforms, scent generation requires a physically manufacturable specification: a recipe composed from a finite palette of ingredients, structured into top, middle, and base notes, with constrained mixture weights. In this talk, I will introduce our recent work on manufacturable text-to-scent generation, which combines large language models for semantic interpretation with a rule-based formulation engine that enforces physical and compositional constraints. I will also briefly discuss ongoing perceptual evaluation results based on 100 manufactured scents and 1,000 human evaluations, which suggest that generated scents can carry weak but measurable traces of their originating textual meanings. Finally, I will outline how manufacturable text-to-scent generation can become a building block for future multisensory holodecks and olfactory interfaces.

Bio

Jooyoung Kim is an Assistant Professor in the Department of Convergence Software at Myongji University, South Korea. He received his B.S. and Ph.D. degrees in Electrical and Electronic Engineering from Yonsei University. Before joining Myongji University, he worked as a Manager in the Data Science Team, G.AI Group at SK Inc. C&C. He also co-founded PS Analytics, an esports data analytics company. His research focuses on applied AI, multimodal generation, olfactory AI, game and interactive AI, biometrics, and LLM evaluation, with an emphasis on applying and developing pattern recognition and machine learning methods across multidisciplinary domains.

© 2023 The International Conference on Holodecks

bottom of page