Generated AI learns to make perfume

The system creates custom smells based on text commands only

Scientists have developed an artificial intelligence system that can create customized perfumes based solely on text descriptions, potentially revolutionizing the traditional labor-intensive perfume industry. Researchers at the Tokyo Institute of Science have launched a breakthrough AI model called Ogdiffusion, which can develop essential oil recipes to match specific odor descriptions such as “woody”, “citrus” or “flower.”

The system, detailed in a study published last month in an IEEE visit, represents the first successful application of a generative diffusion network, an AI type commonly used in image and text creation for the field of aroma design.

“Our diffusion network uses patterns in mass spectral data of essential oils to generate new perfume profiles in fully automatic, simplified and data-driven methods while maintaining high-quality data output,” explained the study’s lead author. “By eliminating human intervention and molecular synthesis from this process, we provide a fast, general and effective method for the generation of fragrances.”

Unlike traditional perfume creations, this depends heavily on trial and error by skilled perfumers, while the Ogdiffusion system can analyze chemical profiles of essential oils marked with 9 common odor descriptors. When the user specifies the desired odor characteristics, the AI ​​generates the corresponding chemical profile and calculates the exact mixture of essential oils needed to recreate the odor.

The researchers verified their system through sensory tests in which participants consistently identified the AI-generated perfume as matching their expected description. In one experiment, participants were able to reliably distinguish odors designed with specific descriptors from odors without specific descriptions.

“This approach represents a significant advance in aroma design,” Nakamoto said. “By automating the production of mass spectra corresponding to the desired odor profile, the Ogdiffusion network provides a more efficient and scalable approach to perfume creation. In addition, even newbies can create the expected odor to create scent digital content.”

Although Givaudan’s Carto, IBM’s Philyra and Firmenich’s Scentmate have assisted with the use of commercial AI tools for perfume formulas, they still require expert guidance and rely on proprietary datasets. By contrast, the Ogdiffusion system can run completely autonomously and produce recipes using easy-to-use essential oils.

This technology may have far-reaching implications, beyond perfumes, and may transform the experience of food seasonings, household products, and even virtual reality. With the development of digital smell, this study demonstrates a future that makes the creation of scents as accessible and customizable as today’s digital images.

Related