Virtual Makeup Try-On for Foundation: How to Settle the Shade Question Before the Purchase

Get a lipstick wrong and the shopper changes her mind. Get a foundation wrong and the shopper gives up. In practice, no beauty category carries as much online purchase anxiety as foundation: dozens of shades with names that mean nothing, swatch photos taken under different lighting and a real fear of receiving a product that does not match the skin. A well-built virtual makeup try-on for foundation attacks exactly that point — it shows the shade applied to the face of the person buying, not on a generic model.

This article is about how to structure that technically: what virtual try-on solves on its own, what it does not solve, and how to combine shade rendering with skin data to turn "what is my shade?" into an objective answer.

Why foundation is the hardest case (and the one with the highest return)

A lipstick works by contrast. A foundation works by matching: it needs to disappear into the skin. That creates three problems that directly affect your operation.

Shade names communicate nothing

"Beige 30", "W4", "Medium neutral". For someone who already uses the brand, it works. For someone new, it is noise. The shopper opens the shade grid, does not recognise her own shade in any name and closes the tab.

A photographed swatch is not a swatch on skin

The product photo was taken in a studio, under controlled light, on someone's arm. The shopper is in her bathroom at home, under yellow light. The comparison is impossible and, worse, it leads to mistakes.

Undertone is invisible to the shopper

Many people get the depth of the shade right and the undertone wrong — they choose a warm shade when the skin leans cool, and the result looks grey or orange. Undertone is a piece of information shoppers rarely know how to report about themselves.

The three problems share the same root: the lack of a visual reference applied to the person's real skin. That is where virtual try-on comes in.

How to build the virtual makeup try-on for foundation

MaIA Try-On delivers this experience in two forms, and the choice between them depends on the channel.

Photo API: the simplest route to integrate

With the API, the flow is direct: you send the shopper's photo and the product — lipstick or foundation — and receive back the image with the shade rendered. The palette is calibrated to your brand's real products within a project, meaning the rendered "Beige 30" is your Beige 30, not a generic colour approximation.

This route works well wherever there is no active camera: WhatsApp, email, customer service chat, a tablet at the store counter, a launch campaign. The shopper sends a selfie, receives the image with the foundation applied, and the conversation continues from there.

Browser SDK: live mode and Full Look

The browser SDK, in early access since September 2026, runs thirteen products — including foundation — plus the combined Full Look, in live mode through the camera (rendered locally, with no image upload) or in photo mode, with any colour in #RRGGBB, hosted by the brand itself. It is the route for the product page and for the site's digital fitting room, when you want the person to turn her face and see the shade in motion.

The distinction matters: the eleven additional products and the Full Look are exclusive to the SDK and work as a local preview. The API renders lipstick and foundation. If your question is which of the two routes fits your case, the comparison is detailed in photo-based virtual try-on vs. other approaches.

Virtual try-on alone does not choose the shade

Here is the part many implementations forget. Virtual try-on shows how a shade looks. It does not say which shade to test first. If you have 28 foundation shades and the shopper has to go through all 28 to find out, you have only moved the doubt somewhere else.

What solves it is entering the experience with a starting point. MaIA Scan analyses the selfie and returns a beauty profile in JSON with skin tone, undertone, skin type and visible skin condition, hair attributes and a confidence level per field. The model was trained on Brazilian skin, which matters precisely in the medium tone range and in undertone, where generic grids tend to slip.

With that profile, the flow changes shape:

1. The shopper sends a selfie.
2. MaIA Scan returns tone and undertone, with confidence per field.
3. You map that profile to two or three candidate shades from your catalog.
4. Try-On renders those candidates on the shopper's own photo.
5. The person chooses between options that already make sense for her.

It is the difference between a fitting room and a consultant. And when the recommendation needs to talk to the whole catalog — foundation, powder, concealer, finish, coverage — MaIA Beauty Advisor plays that role on WhatsApp, on the product page and in chat, reading the beauty profile and suggesting the brand's own products. It is delivered as a project with the B4A team, not as a self-serve API.

What to consider before going live

Calibrate the palette carefully. The quality of a foundation virtual try-on depends on the rendered shade matching the physical product. That calibration is done in the project, product by product.

Define what to do with low confidence. The profile carries confidence per field. If the photo has poor light, the interface should ask for another selfie or widen the range of suggested shades, instead of forcing an answer.

Do not oversell what the analysis does. The analysis describes visible attributes of a photo. It is not a clinical assessment and does not replace a health professional.

Keep the data. Every analysed selfie becomes first-party data: you start to know the real distribution of tones and undertones in your customer base — direct input for assortment, stock and the brief for upcoming launches.

Create your account, purchase credits and get your API key in the MaIA documentation — or talk to the team to calibrate the palette of your foundation line.