AI can now create custom fragrances from scent labels

Scent plays a vital role in how we experience the world. It shapes our memories, emotions, and choices. Fragrance companies spend years perfecting the right blend. But what if artificial intelligence (AI) could do it in minutes?

A team from the Institute of Science Tokyo has developed an AI model that does just that. The system, named Odor Generative Diffusion (OGDiffusion), uses essential oil data and scent descriptors to create new, ready-to-blend fragrances.

The work signals a shift from traditional perfume artistry to data-guided aroma creation.

AI overcomes obstacles in scent design

In perfumery, food, and household products, scent influences our reactions. Traditionally, only experienced perfumers could design blends. The process involves many trial-and-error steps, often leading to inconsistent results.

OGDiffusion removes this bottleneck. It uses mass spectrometry data from essential oils as a chemical fingerprint. This allows the model to understand and recreate scents without needing chemical knowledge or manual testing.

The innovation lies in its learning method. The model starts with noisy data and learns to clean it up, mapping it to scent labels like “citrus” or “woody.” It then builds a scent profile using essential oils and a mathematical method called non-negative least squares.

Training AI to create scents

Commercial systems like IBM’s Philyra and Firmenich’s Scentmate assist perfumers. However, they depend on private datasets and still need human input. OGDiffusion is different. It uses open data, works without expert help, and creates blendable recipes using essential oils.

“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. By eliminating human intervention and molecular synthesis from the process, we provide a fast, general, and efficient method for fragrance generation,” explained Nakamoto.

Unlike other AI systems, this model was trained on data from 166 essential oils and 9 common odor descriptors. The training process involved adding controlled noise to the data, then teaching the network to reconstruct the original scent information.

AI creates scents with precision

Creating data-driven scents is one thing. Matching human expectations is another. The researchers performed several sensory tests with human volunteers to validate their work.

In one test, participants had to match AI-generated scents to their correct descriptors. In another, they had to tell apart blends with and without a specific scent label. Both tests showed high accuracy. Volunteers could reliably identify the correct scent or pick the one that matched the descriptor.

A third test asked participants to rank AI-generated scents and real essential oils by strength of a single descriptor. The AI blends consistently ranked higher. This confirmed the system’s precision in producing distinct and expected scent profiles.

An efficient and scalable method

“This approach represents a significant advancement in aroma design,” said 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.”

The OGDiffusion model bridges chemistry and sensory science. It understands how changes in chemical composition affect what we smell. It can even generate multiple variations of the same scent by using different input noise, introducing creative diversity into the process.

Aroma by algorithm: What’s next

The researchers see potential far beyond the lab. OGDiffusion could revolutionize how fragrances are designed in perfumery, flavoring, cosmetics, and even virtual reality.

Scented digital content is already a reality in experimental spaces. This AI system could make scent design scalable and accessible.

There are limits, though. The current model works with only nine odor descriptors. Adding more would require a larger training set and better data curation. Also, standardizing scent language across datasets remains a hurdle.

Still, the network can expand. It can potentially integrate natural language inputs using embedding tools like fastText. That would allow users to describe a scent in plain words – and get a fragrance back.

AI can now customize scents

The model proves that scent creation can be automated, validated, and customized. By using mass spectrometry data as input and blending essential oils as output, OGDiffusion creates a closed loop of design and realization.

The vision is clear: computers not only see and hear – they can now “smell” in a meaningful way. And soon, you might design your own perfume with a few words and clicks. That’s more than a technical win – it’s a sensory revolution.

The study is published in the journal IEEE Access.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–