AromaGen: Interactive Generation of Rich Olfactory Experiences with Multimodal Language Models
Researchers' neck-worn device creates custom food aromas using 12 base odorants and an LLM.
A research team from MIT and Carnegie Mellon University has developed AromaGen, a breakthrough AI system that generates real-time olfactory experiences from free-form text or visual inputs. The technology represents a significant leap from traditional scent interfaces limited to fixed cartridges, using a multimodal language model to leverage latent olfactory knowledge. The system maps semantic inputs to structured mixtures of 12 carefully selected base odorants, which are then released through a neck-worn dispenser.
Users can interactively refine generated aromas through natural language feedback, with the system employing in-context learning to improve results. In a controlled study with 26 participants, AromaGen matched human-composed mixtures in zero-shot generation and significantly surpassed them after iterative refinement. The system achieved a median similarity rating of 8/10 to real food aromas and reduced perceived artificiality to levels comparable to actual food.
The technology opens new possibilities for communication, wellbeing applications, and immersive technologies by bringing dynamic olfaction into interactive systems. By overcoming the limitations of large-scale olfactory dataset scarcity, AromaGen demonstrates how AI can bridge the gap between digital interfaces and sensory experiences, potentially transforming fields from entertainment to therapeutic applications.
- Uses multimodal LLM to map text/images to mixtures of 12 base odorants
- Achieved 8/10 similarity to real food aromas in user study with 26 participants
- Enables iterative refinement through natural language feedback via in-context learning
Why It Matters
Brings dynamic scent generation to interactive systems, enabling new applications in communication, wellbeing, and immersive experiences.