Generative AI learns to make perfume

System Creates Custom Scents Solely From Text Commands

Scientists have developed an artificial intelligence system capable of creating custom fragrances based solely on text descriptions, potentially revolutionizing the traditionally labor-intensive perfume industry. Researchers at the Institute of Science Tokyo unveiled a breakthrough AI model called Odor Generative Diffusion (OGDiffusion) that can formulate essential oil recipes to match specific scent descriptions like “woody,” “citrusy,” or “floral.”

The system, detailed in a study published last month in IEEE Access, represents the first successful application of generative diffusion networks—a type of AI commonly used for image and text creation—to the realm of scent design.

“Our diffusion network uses patterns in mass spectrometry data of essential oils to generate new fragrance profiles in a fully automated, streamlined, and data-driven approach while maintaining high-quality data output,” explains Professor Takamichi Nakamoto, the study’s lead author. “By eliminating human intervention and molecular synthesis from the process, we provide a fast, general, and efficient method for fragrance generation.”

Unlike traditional perfume creation, which relies heavily on trial-and-error by skilled perfumers, the OGDiffusion system works by analyzing the chemical profiles of 166 essential oils labeled with nine common odor descriptors. When users specify desired scent characteristics, the AI generates a corresponding chemical profile and calculates the precise mixture of essential oils needed to recreate that scent.

The researchers validated their system through sensory tests in which participants consistently identified the AI-generated fragrances as matching their intended descriptions. In one experiment, participants could reliably distinguish between scents designed with specific descriptors versus those without.

“This approach represents a significant advancement in aroma design,” states Nakamoto. “By automating the generation of mass spectra corresponding to desired odor profiles, the OGDiffusion network offers a more efficient and scalable method for fragrance creation. Moreover, even a novice can create an intended scent to make scented digital contents.”

While commercial AI tools like Givaudan’s Carto, IBM’s Philyra, and Firmenich’s Scentmate already assist with fragrance formulation, they still require expert guidance and rely on proprietary datasets. In contrast, the OGDiffusion system operates completely autonomously and produces recipes using readily available essential oils.

The technology could have far-reaching implications beyond perfumery, potentially transforming food flavoring, household products, and even virtual reality experiences. As digital olfaction advances, this research suggests a future where scent creation becomes as accessible and customizable as digital imagery is today.

Related