Built like a laboratory.
Not a guessing game
Every formula Fragrance Engine generates is reproducible. The same inputs always produce the same output, because the engine is physics-backed and deterministic, not an LLM dressing up perfumery vocabulary.
Describe
Tell it what you're after in plain words, or dial in descriptors and an archetype yourself.
Draft
The engine composes a balanced formula from real materials, with a full top-to-base structure you can see.
Refine
Edit it gram by gram, and the pyramid, the projection, the cost, and the lab math all recalculate as you work.
Source
Pull live supplier prices onto the formula, or swap in a cheaper material and let the engine rebalance so the accord still holds.
Export
Print a lab-ready sheet with exact masses for any batch size, ready for the bench or the manufacturer.
What makes the engine trustworthy
Plenty of tools promise a "fragrance AI" and stop there. The harder question is what actually happens between your brief and the percentages on the finished lab sheet, and the answer comes down to four things that run in order.
Deterministic engine
A 15-stage pipeline transforms scent descriptors into balanced formulas. No randomness. No black box. Run it twice with the same brief, you'll get the same formula twice, to the percentage.
Physics-backed evaporation
Every material's behavior is computed from real molecular data: boiling point, molecular weight, vapor pressure. The same Clausius–Clapeyron framework professional labs use to predict volatility.
Density-aware lab math
Percentages are by mass, but you measure by volume. The engine computes the mass-weighted density of every formula and converts mL to grams correctly, so your scale matches your spec.
Every formula, read back to you
The moment a formula exists, the engine reads it: the top, heart and base pyramid, projection, longevity, structure, and cost. The bench analysis a perfumer does by hand, done for you on everything.
One brief, one formula
Generative AI tools roll the dice every time you press the button. Useful for inspiration, hopeless for actual lab work. The Fragrance Engine pipeline is deterministic: descriptor weights → semantic match against the materials library → physics-derived role assignment → constraint solver → normalized percentages. Same inputs, same outputs, every time.
That means you can save a formula, hand it to a chemist or contract manufacturer, and reproduce it weeks later without explaining a prompt.
"If you can't run the formula twice and get the same liquid, it's not a formula. It's a sketch."
Boiling points and molecular weights, not vibes
Whether a material reads as a top note (bright, fleeting) or a base note (warm, persistent) is not an opinion. It's a function of boiling point, molecular weight, and vapor pressure. The engine reads those three values for every material in your library and computes a volatility index, a number between 0 (a heavy resin) and 1 (a flash-off citrus).
The 60/40 BP-vs-MW weighting comes from Calkin & Jellinek's "Perfumery: Practice and Principles", the same textbook taught at Givaudan's perfumery school. Boiling point is the dominant predictor of room-temperature evaporation rate, with molecular weight as a correction.
From that single volatility number, the engine derives every other behavioral metric: projection (how far it throws on a first spray), longevity (how long it lingers on skin), sillage (the trail you leave behind). Each one is a transparent formula, documented openly in the engine reference for the curious or skeptical.
Mass-weighted density
Most online "perfume calculators" pretend 1 mL = 1 g. It's not. A citrus-heavy top is closer to 0.88 g/mL. A vanillin-laden base can reach 1.05 g/mL. At 10 mL of concentrate, that's a 1.7 g difference between what your scale reads and what your spec sheet says.
Standard physical chemistry. The engine reads each material's actual density from a curated catalog, computes the mix density for your formula, and prints real per-material gram amounts on the lab sheet. When the mix density deviates from water by more than 0.05 g/mL, a flag appears so you know the volume and weight figures disagree.
Two catalogs. Each what it's best at
The engine pulls per-material physical data from two complementary sources, each curated for what it does best. Where they disagree, a documented per-field priority decides the winner.
PubChem (NIH)
The U.S. National Library of Medicine's reference database. Molecular weight, boiling point, vapor pressure, logP, all lab-measured or computed against a peer-reviewed standard. Trusted by every academic chemistry program on earth.
The Good Scents Company
The de-facto industry catalog of fragrance materials: density, recommended use levels, scent descriptions, and supplier-verified physical data for the naturals and captives PubChem doesn't cover.
For your own laboratory inventory, you can also save lot-specific overrides, a Rose Otto from one supplier may measure differently from another. The engine prefers your measured values over the catalog defaults when it computes a lab sheet for that lot.
What it costs, and where to buy it
Every formula carries a live cost per gram, pulled from real supplier listings where we have them and the catalog price otherwise. Each material links out to where you can buy it, so a formula on screen doubles as a shopping list for the bench.
The cost optimizer goes further: it finds descriptor-matched substitutions that bring the price down while the olfactive shape barely moves.
An honest asterisk
This is a physics-informed approximation. Not a wet-lab measurement
Real fragrance performance depends on skin chemistry, ambient temperature, humidity, application method, concentration, oxidative aging of ingredients, and how the wearer's clothes interact with the alcohol carrier. The engine's projection / longevity / sillage hours are useful for comparing two formulas against each other, not for predicting an exact number on a specific human. Where lab data exists for real perfumes (Aventus, Sauvage), engine estimates land within roughly ±1.5 hours on longevity and within one bracket on projection. Useful, accurate enough for design, and we say so out loud.
Ready to compose?
Start a 7-day trialThe full engine and the real 260-material catalog, free to try for a week.