AromaGen: AI Wearable Generates Scents from Text and Images

Researchers have developed AromaGen, an AI-powered wearable interface that generates real-time aromas from free-form text or visual inputs. The system uses a multimodal large language model to map semantic inputs to structured mixtures of 12 base odorants, released through a neck-worn dispenser. Users can refine scents through natural language feedback via in-context learning. In a controlled user study (N=26), AromaGen matched human-composed mixtures in zero-shot generation and significantly surpassed them in iterative refinement. The work addresses the limitations of current olfactory interfaces, which rely on fixed scent cartridges and predefined patterns, and overcomes the scarcity of large-scale olfactory datasets by leveraging latent olfactory knowledge in LLMs. The paper is published on arXiv (2604.01650).