Frequently asked questions.

Conventional try-on matches pixels to pixels. It compares what your camera produced against what a studio camera produced, and assumes both were lit by neutral daylight. That assumption falls over in most real rooms, and it falls over hardest on deeper skin tones, where sensor noise and channel clipping do the most damage.

ExactHue Core models the lighting instead of assuming it. It estimates the illuminant, inverts the camera's processing chain, recovers an estimate of the skin's own spectral reflectance, and matches in CAM16-UCS — a space where numerical distance tracks perceived difference. Every result carries an explicit ΔE₀₀ and a calibrated confidence, so the answer is auditable rather than merely assertive.

None, at any stage. Texture is signal — it's the evidence that a match is behaving on real skin. Brightening shifts the very thing we're trying to measure, and smoothing throws away the specular detail that makes a preview believable in the first place. A flattering preview is the fastest route to a return, so beautification isn't just scientifically wrong here, it's bad business for the brand paying us.

The engine declines. Captures are qualified before analysis for exposure headroom, per-channel clipping, motion blur and the presence of more than one illuminant. If the capture cannot support a reliable answer, the response is a structured reason code rather than a recommendation, so your interface can ask the customer to move nearer a window or turn off a mixed light source.

This makes demos less impressive and production far more trustworthy. Calls rejected at qualification are not billed.

Accuracy is reported per Fitzpatrick band and per undertone family, never as a single headline average. A single average would let us hide precisely the failure that has damaged trust in this category for a decade.

The shade library is balanced by sampling design across Fitzpatrick types I–VI, with undertone families stratified within each depth band. A regression in any individual band blocks a release, even when the aggregate improves.

Three things, and we'd rather you hear them from us. Heavy mixed lighting — daylight down one side of the face, a warm bulb on the other — is still the hardest common case, and today we detect it and decline rather than solve it. Very low light degrades reflectance recovery past the point where we'd trust the answer. And metallic, duochrome and heavily pearlescent finishes aren't well described by a single measured colour at all, so we're noticeably less accurate on those than on complexion products.

We raise all three in technical calls rather than discovering them with you in production.

Complexion products are the core case: foundation, concealer, tinted moisturiser, tinted serum, cushion, powder and colour-correcting bases. Lip and cheek products are supported with good results for cream and satin finishes. Heavily metallic and duochrome finishes are supported with a stated accuracy caveat, as above.

The technical work is usually an afternoon. One REST endpoint returns ranked matches as plain JSON; you either drop in our embeddable widget or render the response with your own components.

The longer step is catalogue measurement — typically about two weeks, depending on range size and whether you already hold spectrophotometer readings. If you do not, you can ship physical samples to our laboratory in Singapore and we will measure the range for you.

Shopify via an app-embed block, Salesforce Commerce Cloud via a cartridge with SFRA hooks, and commercetools via an API extension. Custom storefronts integrate directly against the REST API with an optional framework-agnostic web component. There is no replatforming requirement and no rebuild of your product page.

The capture step needs a camera and therefore needs JavaScript. Everything around it does not: the widget degrades to a server-rendered shade grid, so your product pages stay complete, readable and indexable for crawlers and for visitors with scripting disabled. This website is built on the same principle — every page here is static HTML.

Seasonal re-measurement is included on Growth and Enterprise plans. A catalogue that drifts out of date quietly degrades every match made against it, so we would rather absorb that cost than watch accuracy decay. New SKUs can be ingested at any time through PUT /v1/catalogue, with validation warnings raised for out-of-gamut or unmeasured entries before they go live.

Growth carries a 99.9% uptime commitment with business-hours support and a named implementation engineer. Enterprise agreements carry a custom SLA with contractual response times. Pilots run on a best-effort basis with a shared support channel. Details are on the service levels page.

It is processed to produce the match and then discarded. Captures are not retained for model training unless the brand has explicitly opted in under a separate written agreement, and ExactHue does not use captures to identify anyone. Processing can be pinned to EU, UK, US or APAC regions for data residency. The full detail is in the privacy policy.

ExactHue Core does not perform facial recognition, identification or verification, and it does not generate a face template. It measures colour from skin regions. That said, image data of a person can be sensitive under several regimes, so we treat captures as personal data, minimise retention, and provide a data-processing agreement covering the arrangement. Where your jurisdiction requires specific consent for this processing, your consent flow governs — we provide the technical controls to honour it.

From controlled research sessions where contributors were informed of the purpose, compensated for their time, and free to withdraw afterwards. Withdrawal removes the sample from the library and from training. No scraping, and no repurposing of customer photographs from brand deployments. The library is used to build and validate the engine; it is not sold, licensed or shared.

Yes — encryption in transit and at rest, least-privilege access, scoped credentials per environment, audit logging on Enterprise, and a responsible disclosure policy. As a pre-seed company we are transparent about what is in place today versus what is on the roadmap; see the security page for the honest version rather than a badge wall.

No. ExactHue Core is a business-to-business engine licensed to beauty brands, retailers and marketplaces. There is no consumer app and no direct-to-consumer product. If you encountered ExactHue through a brand's website, that brand is the party you have a relationship with.

Because it involves real laboratory work: measuring up to sixty SKUs and analysing a year of return data is several weeks of effort. A free pilot would create pressure to cut exactly the corners that make the read-out meaningful. The fee is credited in full against your first twelve months if you continue. See pricing.

Fair question, and better asked now than in year two. Three practical answers. Your measured catalogue is yours — exportable in an open format whenever you want it, including on the day you leave, so the most valuable thing the engagement produces walks out with you. The integration is a single endpoint, so pulling it out costs about what putting it in did. And we'd rather tell you a pilot missed your baseline than book revenue that churns in six months. Judge us on the method and the read-out, not the funding stage.

Yes. We will walk your engineering and science teams through the benchmark construction, the per-band results, and the failure cases, under mutual NDA if you prefer. Book a technical call.