Fragrance design, long an art shaped by intuition and trial-and-error, is entering a new era powered by artificial intelligence.
Researchers at the Institute of Science Tokyo have developed an AI model called Odor Generative Diffusion (OGDiffusion) that creates custom fragrances based on user-defined scent descriptors. The system, built using a generative diffusion neural network, analyzes the chemical profiles of 166 essential oils and links them to nine scent categories such as “citrus,” “woody,” and “floral.”
When a user inputs a desired scent description, the model generates a corresponding chemical profile and calculates the ideal blend of essential oils to match it, eliminating the need for expert guidance or repetitive testing.
Unlike previous AI fragrance tools that rely on proprietary datasets and human guidance, OGDiffusion is fully automated and based on open-source information. Researchers say it marks a major step forward in scalable, data-driven aroma creation.
To evaluate its accuracy, the research team conducted double-blind sensory tests. Participants consistently identified the AI-generated scents with the correct descriptors and, in some cases, rated the synthetic blends as more distinct than real essential oil samples.
“This is a new frontier for scent creation,” said Professor Takamichi Nakamoto, who led the project. He envisions applications ranging from perfumes and cosmetics to immersive technologies like virtual reality.
Currently, the system uses a limited number of scent labels, but the team is working to expand its capabilities. With OGDiffusion, designing a fragrance may soon be as simple as typing a sentence. The era of algorithmic scent has arrived, offering a novel way to experience and shape the sensory world.