OneDress
Six bridesmaids. Six skin tones. One dress color — provably no one's worst option.
The problem
One color. Final sale.
Chosen by eyeballing a stranger.
One color for every complexion
4–8 women, often spanning Fitzpatrick I–VI, wear the same mandated colorway. The shade that flatters the bride can drain everyone else — in the photos that last forever.
Usually non-refundable
Ordered months out, frequently final-sale. The color decision is effectively irreversible — there is no "return it if it washes her out."
Decided by guesswork
Picked from one retailer model who matches nobody in the party. Mailed swatch kits and the "mismatched dresses" trend exist because one-color-for-all keeps failing.
The insight
We inverted the personal-color quiz.
"What are my best colors?"
One selfie → one verdict. Run it for five bridesmaids, get five incompatible answers — which chooses zero dresses. Single-user verdicts don't compose into a group constraint.
"Given one garment all N must wear — which color harms the group least?"
A fixed garment constraint × N measured complexions → one answer. A constrained group-optimization problem no single-user tool can express.
It's a max-of-minimum — Rawlsian — fairness problem. The winner is never any individual's top pick, and by construction the deepest-skin bridesmaid is the protected one, not the sacrificed one.
How it works
Measure. Score. Render. Finish.
Measure
Two photos per bridesmaid. Live analyses return her real skin hex, Fitzpatrick I–VI, and face shape.
Score
24 colorways × N profiles, scored by a published formula. Pure math on cached measurements — zero API units to re-score.
Render
The maximin winner rendered photorealistically on every bridesmaid's own photo — fidelity ΔE00 median 7.8, range 5.5–11.2.
Finish
Face shape picks the earring silhouette; undertone picks the metal — gold warm, silver cool — chained onto the render.
The math
Averages hide victims. Minimums protect them.
flatter(p,c) = 0.50·U + 0.30·C + 0.20·S U undertone harmony C value contrast S saturation harmony groupScore(c) = minₚ flatter(p,c) winner = argmax_c groupScore(c) by-eye = argmax_c mean flatter — how it's chosen today
The by-eye pick
maximize the averageaverage 66 · worst bridesmaid 39 — Type VI pays for the pretty average
The OneDress pick
maximize the minimumaverage 65 · worst bridesmaid 58 — one average point traded to lift the party floor by +19
Real engine output for the six synthetic Fitzpatrick I–VI profiles in lib/demo/demo-party.ts. The engine picks marigold with nobody below 57.8; the mean-maximizing pick (rust) drops the deepest-skin member to 38.8. Two ways to read the gain, both real: the party floor rises +19.1 (38.8 → 57.8), and the worst-off person under rust rises +26.5 (38.8 → 65.3) — different metrics, not the same number. Treat the size of both as illustrative: a units bug in our own undertone term inflates them, published in full at docs/scoring-defect.md rather than patched. Derivation: lib/colorway/engine.ts.
Verified live
Not a mock. Here's the receipt.
| Endpoint | Role | Verified live result |
|---|---|---|
| skin-tone-analysis | measured skin hex — the scoring input | #bb9982 → ITA° 43 · hue° 61 |
| fitzpatrick-scale-analyzer | depth I–VI badge — displayed, never scored | Type II returned |
| face-attr-analysis | face shape → earring silhouette | "Heart" returned |
| cloth-v3 | winner rendered on each bridesmaid | photoreal · ΔE00 median 7.8 (5.5–11.2) |
| 2d-vto/earring | chained finishing touch | gold hoop landed on the render |
Sponsor fit
Topic C, actually combined — Skin AI drives the VTO.
Skin AI The measurement half
- skin-tone-analysis — the hex every score is computed from
- fitzpatrick-scale-analyzer — the inclusivity axis, I–VI
- face-attr-analysis — face shape steers the finish
Apparel VTO The proof half
- cloth-v3 — the winner on every real bridesmaid
- 2d-vto/earring — chained; undertone picks the metal
Deliberately beyond the sponsor's own surface: Perfect Corp ships personal-color analysis — "what's my season?" for one user. OneDress answers the strictly harder inverse for N people at once. Remove any one endpoint and the flow visibly breaks — the API is the engine, not decoration.
Market & impact
A billion-dollar decision, made by eyeballing one photo.
And the beachhead generalizes: prom groups, quinceañera courts, uniforms, ensembles — every "one garment, many bodies" decision. The maximin objective means the darkest-skin member is protected by construction — inclusivity as an algorithm, not a tagline.
Figures are industry estimates (The Knot Real Weddings Study, retailer pricing); swatch-kit sales and the mismatched-dress trend are the revealed-preference proof the pain is real.
Roadmap
Engine proven. Product shipped.
Done & reproducible
- Live 5-endpoint integration — npm run spike
- Typed, Zod-validated client — rate-limited, retrying
- Published scoring engine — 82 passing tests
- Full CI harness — lint, tests, CodeQL, secret scan
- 7-step product UI — Create → Verdict, live at /party
- Cached zero-unit demo party — instant for judges
- Counterfactual Compare screen — by-eye vs maximin
In build → Aug 17
- ≤3-min demo video — live browser, real calls
- Live-run polish — bring-your-own-key, metered end to end
The ask: judge the mechanism on today's receipts — the inversion, the published math, the five live endpoints. The interactive party lands before the deadline, on this same engine.
Six bridesmaids. One color. Zero arguments.
OneDress — the max-of-minimum dress-color engine, built on the YouCam API.
onedress.edycu.dev · YouCam API Hackathon (Devpost) · Topic C
OneDress
← → to navigate · P notes · O overview