MistMaker and Large Language Model MistMaker(LLMMM): LLM Operated Mist Making

Abstract

Ultrasonic mist-making devices are commonly deployed as closed, appliance-level systems with limited programmability. At the same time, recent multimodal large language models (LLMs) have expanded interaction beyond text to visual and auditory modalities, while environmental and atmospheric outputs remain largely unexplored. This paper introduces MistMaker, an open-source, programmable ultrasonic mist platform, and Large Language Model MistMaker (LLMMM), a system that connects LLMs to mist generation through an API-mediated, interpretive control pipeline. Through two hardware pilots, we examine how mist is perceived, interpreted, and appropriated in social and exploratory settings, informing LLMMM's interaction framing and highlighting mist's ambiguity as a nonverbal, non-symbolic medium. We contribute an open hardware infrastructure for programmable mist, a technical architecture for LLM-guided environmental interaction, and an interaction framing that extends multimodal AI beyond representational media toward material, experiential, and atmospheric modulation. This work expands HCI discussions on AI agency, ambient interaction, and environmental media in LLM-mediated systems.