Beauty Profile From a Selfie: The Fields, the Confidence Scores and What to Build With Them

Most beauty personalization projects stall at the same point: the team agrees that customers should see the right shade and the right routine, but nobody can say what data that decision would be based on. A beauty profile from a selfie closes that gap. One photo goes in, and structured, machine-readable attributes come out — skin tone and undertone, skin type and condition, hair attributes — each with a confidence score attached. From there it is an engineering problem, not a guessing game.

This article walks through what those fields are, why the confidence score matters more than most teams expect, and where a profile actually earns its keep once it exists.

What a beauty profile from a selfie actually contains

MaIA Scan is delivered as an API: your application sends a photo, and the response is JSON. Nothing about the output is meant to be shown raw to a shopper — it is meant to be consumed by your own logic, your own copy and your own catalog.

Skin attributes

The skin side of the profile covers tone and undertone, plus skin type and visible condition. Tone and undertone are the two fields that shade matching depends on: tone tells you roughly where a foundation sits on a depth scale, undertone tells you which direction to lean within that depth. Skin type and condition describe what is visible in the photo at that moment, which is what drives routine and category recommendations.

Hair attributes

Hair attributes extend the same idea to a category most personalization projects ignore. If your catalog spans skincare, makeup and hair, a single scan can feed all three instead of forcing the customer through three separate quizzes.

A confidence score per field

Every field comes back with its own confidence. This is the part that changes how you design the experience. A profile is not a single yes-or-no answer; it is a set of independent readings, some of them stronger than others depending on lighting, framing and image quality.

Why the confidence score should shape your UX

Treat confidence as a routing signal, not as a footnote.

High confidence on a field means you can act on it directly: pre-select the shade, filter the grid, personalize the headline.

Low confidence on a field means you fall back. Show a short range of adjacent shades instead of one. Ask a single clarifying question. Invite a second photo in better light. What you should not do is hide the uncertainty and let the customer discover it at delivery, because a wrong shade that felt confident is worse for returns and trust than a shortlist that felt honest.

This also gives your analytics team something useful. If confidence on a given field is consistently low for a segment of traffic, that usually points at a capture problem — a dark in-store lighting setup, a cropped selfie frame, a mobile web flow that compresses images too aggressively — and those are fixable.

Three places a beauty profile earns its keep

On the product page

The profile is what makes a shade recommendation defensible. Instead of a generic swatch grid, the customer sees a shortlist that reflects their own tone and undertone. Pair it with MaIA Try-On and the recommendation becomes visual: the photo-based API renders a lipstick or a foundation from your real palette onto the customer's photo and returns the image, so the shortlist is something they can look at rather than something they have to trust.

The browser SDK, in early access since September 2026 and granted per account, goes further for brands that want a self-hosted experience: thirteen products — from lipstick and foundation to eyeliner, bronzer, contour, highlighter and lip liner — plus a combined Full Look, in live camera mode where rendering happens locally in the browser with nothing uploaded, or in photo mode. Those eleven additional products and the Full Look are SDK-only local previews; the API itself handles lipstick and foundation.

In CRM and lifecycle marketing

A beauty profile is durable first-party data. Stored against a customer record, it lets you build audiences that mean something: replenishment timing by product category, re-engagement flows that reference the right shade family, launch campaigns sent only to profiles the new product was designed for. One scan at acquisition keeps paying off for months.

In conversational channels

On WhatsApp or in on-site chat, the profile is the missing context that makes a recommendation feel like advice instead of a menu. MaIA Beauty Advisor reads the beauty profile and recommends products from your own catalog — it is delivered as a project with the B4A team rather than as a self-serve API, precisely because the catalog mapping and the tone of voice are brand-specific work.

Trained on Brazilian skin

A profile is only as useful as the population the model learned from. MaIA Scan is trained on Brazilian skin, which matters in a market where tone and undertone combinations are wide and where generic models tend to compress the middle of the range into a handful of buckets. If your customers are Brazilian, the difference shows up in the undertone field — the one that decides whether a foundation looks right or slightly off.

What a beauty profile is not

It is not a diagnosis. MaIA describes visible attributes of a photo — it does not diagnose, treat or replace a dermatologist or any other professional, and your copy should say so plainly. Keeping that boundary explicit protects the brand and, in practice, makes customers more willing to scan, because the promise is modest and verifiable.

It is also not a one-time asset to be thrown away. Store it, version it, let the customer re-scan when their routine or the season changes.

Getting from zero to a first profile

The integration is deliberately small: create an account, buy credits, get an API key, send a photo, read the JSON. Start with a single surface — one product family on one page — and decide in advance which fields you will act on and what your fallback is when confidence is low. Everything after that is iteration.

Ready to try it? Create your account and API key and get your first beauty profile back today, or talk to the B4A team about a catalog-integrated rollout.