Tackling the complexities involved in perfume formulation, scientists from the Institute of Science Tokyo (IST) have created an artificial intelligence system designed to automatically produce novel fragrances according to specific aroma preferences provided by users. This AI utilizes mass spectrometry data from various essential oils along with their associated smell characteristics to formulate combinations of these oils into innovative scent mixtures.
This advance could be a game-changer for the fragrance industry, moving beyond trial-and-error to enable rapid and scalable fragrance production. The findings are published in
IEEE Access
.
Creating novel scents plays a vital role in sectors such as perfume-making, cuisine, and household goods, since aroma greatly affects how people perceive these items. Nonetheless, conventional methods for crafting fragrances tend to be lengthy processes heavily reliant on the talent and know-how of professional perfumists. This approach usually involves considerable effort and multiple iterations before reaching the intended olfactory outcome.
To streamline this procedure, a group of researchers headed by Professor Takamichi Nakamoto from Science Tokyo created an artificial intelligence system named Odor Generative Diffusion (OGDiffusion). This framework employs generative diffusion models—machine learning algorithms that learn to produce fresh material by retracing a noisy pattern guided by pre-existing information.
These models are already widely employed to generate images and text, and the team has adapted this technology to create new fragrances.
The system functions through examining the chemical compositions (mass spectrometry information) from 166 crucial oils, each identified using one out of nine possible scent descriptions like “citrus” or “woodsy”.
If users provide specific fragrance preferences, the AI creates an associated chemical blueprint (mass spectrum) that matches these descriptions. Subsequently, it determines the blend of essential oils required to replicate that aroma through a calculation technique known as non-negative least squares.
Nakamoto states that their diffusion network leverages patterns found in mass spectrometry data of essential oils to create novel fragrance profiles through an entirely automated, optimized, and evidence-based methodology, ensuring superior quality outputs. This technique omits manual involvement and molecular synthesis, offering a swift, versatile, and effective means of generating fragrances.
Although current AI-driven perfume development models exist, they depend on exclusive databases and still necessitate specialist involvement. The key benefit of this novel approach lies in its capability to fully automatize the production of innovative aromas. Additionally, since the system generates perfumes using essential oil formulas, the resulting scent can be readily reproduced.
Moreover, the team performed human sensory evaluations to determine if the AI-created perfumes matched their designated scent profiles. In a double-blind configuration, 14 individuals were asked to associate AI-produced fragrances with suitable descriptions like “lemony” or “flowery.”
Participants were consistently able to identify the correct fragrance, demonstrating that the system could produce scents that met people’s expectations. In another test, participants distinguished between two scents: one designed to express an additional specific odor descriptor and an original scent without that descriptor.
They consistently chose the fragrance that aligned with the intended description, suggesting that the model produces distinct and recognizable scent profiles.
Nakamoto’s concept—the pioneering one—signals a future where AI revolutionizes fragrance creation. As Nakamoto points out, “This method marks a substantial progress in the field of perfume design.”
Furthermore, he states, “The OGDiffusion network streamlines the process of generating mass spectra for targeted odor profiles, providing a more efficient and scalable approach to fragrance development. Additionally, this makes it possible for beginners to produce their desired scents to generate fragrant digital content.”
To sum up, this cutting-edge technique facilitates quicker and more adaptable fragrance creation, with possible uses spanning multiple sectors. Through the use of artificial intelligence in scent production, the OGDiffusion model shows that machines have the capability to exhibit creative flair when it comes to olfactory perception.
More information:
Manuel Aleixandre et al., Generative Diffusion Network for Crafting Fragrances,
IEEE Access
(2025).
DOI: 10.1109/ACCESS.2025.3555273
Furnished by the Institute of Science Tokyo
This tale was initially released on
Tech Xplore
. Subscribe to our
newsletter
For the most recent updates on science and technology news.
