What is Carto, IFF's AI fragrance tool?

Carto is an AI-assisted composition platform developed by IFF (International Flavors and Fragrances, New York, USA). It allows perfumers to select aromatic building blocks and receive real-time formula suggestions, ingredient substitutions, and IFRA compliance checks. The system learns from the perfumer's choices over time and can generate formulas within specified parameters (family, concentration, regulatory constraints). Carto was publicly announced around 2018-2019 and has been developed further since. Givaudan operates equivalent proprietary tools under different names.

Can AI create a perfume independently of a human perfumer?

Current AI systems in perfumery do not operate independently. They are tools that support the work of human perfumers: suggesting formulas, checking compliance, predicting preference scores. The final evaluation of a formula is always sensory and human-led. An AI system cannot smell: it works with numerical representations of ingredients and their relationships. The creative decision about whether a formula achieves its intended effect, how it evolves over time on skin, and whether it has the right emotional register for a brief is made by a human perfumer evaluating the physical material.

Does AI use change who gets credited for a perfume?

The use of AI tools does not currently change the authorship credit of a fragrance. A perfumer who uses Carto to generate formula variants, then selects and refines one of those variants through sensory evaluation and creative judgment, is still credited as the creator. This is analogous to a composer using digital audio software: the technology accelerates parts of the process without displacing the human creative authorship. However, as AI tools become more capable, questions about authorship and attribution in AI-assisted creation will become more complex.

Are there niche houses that use AI in fragrance development?

Most niche houses that work with major supplier perfumers have indirect exposure to AI tools, because the supplier companies (Givaudan, Firmenich, IFF, Symrise) use AI in their development processes. Independent niche houses working with self-trained or freelance perfumers have variable access to AI tools. Several independent fragrance technology startups have developed AI tools aimed at smaller-scale perfumers and indie houses, though the most sophisticated systems remain at the supplier level.

What is AI olfactive design in perfumery?

Q: What is AI olfactive design in perfumery?

AI olfactive design refers to the application of artificial intelligence tools in fragrance creation. This includes: generating and evaluating thousands of formula variations from a creative brief; predicting consumer preference scores for proposed formulas based on historical data; identifying non-obvious ingredient combinations through pattern recognition in fragrance databases; and accelerating the early stages of development that would previously require weeks of manual work. By 2026, AI tools are in routine use at all major fragrance suppliers but do not replace human sensory evaluation or creative direction.

The essentials

AI olfactive design names the set of machine learning tools that suggest formulas, propose ingredient combinations, predict consumer preference, and accelerate the early stages of fragrance composition. The first commercially documented system was IBM Philyra, presented in 2018 in partnership with Symrise, followed in 2019 by Givaudan Carto, an internal tool for accord exploration (Perfumer & Flavorist, accessed 2026-05-29). The tools work on digital representations of raw materials and their interactions, but the final judgement remains sensory and human.

By 2026, the four leading houses, Givaudan, Firmenich, IFF, and Symrise, have all integrated AI-assisted modules into their internal creation platforms. These systems are used to map formula spaces, propose alternates to restricted materials, and rank options against consumer panel data. They sit upstream of the perfumer rather than replacing them, and most output is filtered, edited, or discarded before any composition reaches a brief presentation.

The technology operates in a constrained domain. There is no general model that generates a complete brief-ready formula on demand, and no system that can evaluate olfactive quality without human noses in the loop. The 2025 wave of synthetic biology platforms, including Osmo founded by Alex Wiltschko in 2023, is extending the field toward predictive olfaction (Bois de Jasmin, accessed 2026-05-29), but commercial use in fine fragrance remains framed around assistance rather than authorship.

Where AI olfactive design came from

Machine learning on fragrance data is older than the current public conversation suggests. Major suppliers were running consumer-preference models on panel data from the early 2010s, drawing on decades of internal accord libraries and structured evaluation records. These projects were unpublicised because the data underlying them was proprietary and the commercial advantage was internal.

The first system designed to generate formulas rather than predict preference was IBM Philyra, announced in October 2018 by IBM Research and Symrise. Philyra trained on 1.7 million formulas and was used to co-create the Symrise fragrance O Boticario Egeo ON me released in Brazil in 2019. Givaudan Carto followed in 2019 as a tablet-based tool that allowed perfumers to explore formula trajectories and accord combinations in real time at the bench.

The reference platforms in 2026

The active commercial platforms in 2026 are concentrated at the four big houses. Givaudan operates Carto and a connected suite for natural and biotech material substitution. Firmenich operates internal generative tools linked to its EmotiCode consumer-research framework. IFF deploys AI modules across its captives library and predictive performance. Symrise continues the Philyra line under its own infrastructure. Smaller suppliers Mane, Robertet, and Takasago use third-party or licensed components rather than full in-house systems.

Outside the established suppliers, the most visible new entrant is Osmo, a 2023 spin-off from Google Research focused on building a digital model of olfaction. Osmo published a 2023 paper in Science on principal odor maps, and by 2026 its platform is being trialled by selected partners in flavour and fragrance. The platform is not yet a generalised creation tool, but it is the first credible attempt at a foundational olfactive model outside the supplier oligopoly.

How a perfumer actually uses these tools

The day-to-day use of AI olfactive design is modest by consumer-tech standards. A perfumer working a brief at Givaudan or Firmenich may query an internal system for accord variations on a starting structure, ask for restricted-material substitutes, or run a candidate formula against predicted consumer scoring before bench evaluation. Each output is one option among many. The perfumer then edits, smells, rebuilds, and iterates with assistants and evaluators across days or weeks.

The tools shorten the speculative phase rather than the creative one. A search for an unexpected accord that once took ten bench trials may be narrowed by a software prompt to three or four. Consumer-panel testing that once required a fortnight of recruitment can be partially simulated against historical data. The composition itself, in the sense of the chosen structure and the final balance, remains a human decision validated by smell.

What AI does not solve

The limits are practical and structural. There is no model that smells. Olfactive performance still has to be evaluated on blotters, then skin, then under varied environmental conditions, and the only judges qualified to make those calls are trained perfumers and evaluators. Consumer-preference models predict patterns observed in past data and are unreliable for genuinely new structures or unfamiliar territories. AI does not invent the brief, the narrative, or the cultural moment a fragrance will inhabit.

There are also regulatory and intellectual-property questions that the industry has not settled. Whether a formula proposed by a generative system can be authored, patented, or claimed by the perfumer is unresolved. The 2024 Senate IFRA-EU working note on AI authorship in fragrance flagged the question for industry consultation, but no binding framework exists in 2026.

Reach into independent niche perfumery

Independent niche houses have limited access to the dominant platforms because Carto, Philyra-derived tools, and the equivalent IFF and Firmenich systems are internal supplier infrastructure rather than commercial software. A niche perfumer working with raw materials from Robertet or Mane may benefit indirectly from supplier-side AI substitutions in captive accords, but most independent creation in 2026 remains traditional bench work with notebooks, weighing scales, and blotters.

A small number of startups offer subscription tools aimed at smaller-scale creators, including Sensorium, Plumesphere, and various open notebooks built on community formula data, but commercial adoption in fine fragrance is marginal. The economic logic of niche perfumery, which values handcraft and a singular signature, runs in the opposite direction of AI-assisted production, and most independent perfumers treat the tools as irrelevant to their practice (Now Smell This, accessed 2026-05-29).

Sources

Perfumer & Flavorist, coverage of IBM Philyra (2018), Givaudan Carto (2019) and supplier AI deployments. Accessed 2026-05-29.
IBM Research and Symrise, Philyra AI fragrance system, technical announcements and case study on Egeo ON me, 2018 to 2019.
Bois de Jasmin, Victoria Frolova, editorial articles on Osmo and computational olfaction. Accessed 2026-05-29.
Now Smell This, editorial pieces on AI in fragrance and independent practice. Accessed 2026-05-29.

Published 29 May 2026 · Updated 30 May 2026 · Last fact check: 30 May 2026 · Osmetheca · Editorial team