Virtual makeup try-on for beauty brands: the customer sends a photo, the product comes back applied

What a photo-based virtual makeup try-on is

A virtual makeup try-on shows a shopper how a product looks on her own face before she buys. In MaIA Try-On that happens from a single photo: the customer sends a front-facing image, the model locates the makeup area and renders the product with the right colour, coverage and finish. Everything that is not makeup — teeth, mouth, skin, facial features — stays exactly as it was. No skin smoothing, no reshaping, no filter.

The site demo shows the thirteen simulators and the Full Look: lipstick in the six launch shades (rose, classic red, berry, coral, plum and rose nude, in matte, satin and gloss) and, for foundation, concealer, powder, bronzer, contour, blush, highlighter, eyeshadow, eyeliner, mascara, eyebrows and lip liner, sample colours with each product's finishes. The same technology is delivered per project with the palette calibrated to your brand's real products. The demo colours are examples, not a brand's catalogue.

The model is B4A's own — trained in-house, with no third-party try-on vendor — on the same beauty foundation that powers MaIA Scan: hundreds of thousands of real selfies from Brazilian consumers, across the full range of Brazilian skin tones. In a try-on that is decisive: the same lipstick has to look true on fair, medium, dark and deep skin.

How it works, in four steps

From photo to result the call is synchronous: the response to the request is the finished image.

1. 1

   The customer sends a front-facing photo

   JPEG or PNG up to 6 MB, shorter side at least 320 px, mouth visible. It can come from an upload, the phone camera or a WhatsApp message. The brand collects the person's consent before sending.

2. 2

   Your application picks shade and finish

   One multipart POST /api/v1/try-on with the photo, the shade and the finish (matte, satin or gloss). The accepted shades come from GET /api/v1/try-on/shades, at no credit cost.

3. 3

   The model renders only the makeup area

   In 2 to 4 seconds the photo comes back as a PNG with the product applied. The X-TryOn-Finish-Requested and X-TryOn-Finish-Applied headers tell you which finish you asked for and which one could be applied to that photo.

4. 4

   You show the result where the sale happens

   On the product page, in the quiz, in the WhatsApp conversation or in an e-mail. The image belongs to the customer: she can save it, compare shades and head to the cart with the right colour.

Who it is for: where a lipstick try-on sells most

Because it starts from a photo rather than a live camera, MaIA Try-On fits channels an augmented-reality mirror cannot enter.

Live through the camera: the MaIA Try-On SDK in early access

Beyond the photo API, MaIA Try-On exists as a browser SDK, in early access since September 2026: thirteen simulators and a Full Look, rendered locally on the customer's device — live through the camera or from a photo — without sending the image to any server.

The SDK is in early access: create your account and ask for access. The integration guide lives in the signed-in area.

Why virtual try-on converts: what brands have already measured

The numbers below come from third parties — brands and vendors with technologies and contexts different from MaIA Try-On. A market reference, not a promise.

What they have in common: virtual try-on removes the "is this shade for me?" doubt from the path to purchase — exactly where MaIA Try-On comes in, on the product page, in the quiz and on WhatsApp, from a single photo.

Your conversion depends on your catalog, your traffic and where the try-on enters the journey. We measure it in your project.

Virtual try-on vs AR try-on: what's the difference?

When someone searches for "AR try-on" or "virtual mirror" they usually picture a live camera tracking the face in real time. MaIA Try-On is not that — and it is worth saying plainly. It is a photo-based virtual try-on: a still image goes in, a rendered image comes out. The two approaches solve different problems.

Live augmented reality needs an SDK on the site or in the app, WebGL or native code, camera permission and a good connection; it shines in store mirrors and in-app experiences. A photo-based try-on only needs an image to exist — which is why it works in WhatsApp, e-mail, the PDP and the quiz with no special front-end integration, produces a file the customer keeps and shares, and lets each shade be calibrated to the brand's real product with far more control over the result.

What the try-on API takes and returns

The MaIA API is an HTTPS gateway: you authenticate with a key from your account (Authorization: Bearer msk_live_…), send multipart/form-data and get the answer in the same request. No webhook, no queue, no session state.

Simulated colour may differ from the actual product. Photos larger than 1600 × 1079 px may be downscaled by the model before rendering; the render service scales to zero, so the first call after an idle period can take longer.

Your brand's palette, not ours

A try-on only sells if the colour on screen matches the product in the box. That is why MaIA Try-On is calibrated per brand: we take your product references — swatches, application photos, each shade's spec — and tune colour, coverage and finish until the render matches the real lipstick across a range of skin tones, with human validation before it enters your project. The same goes for makeup categories beyond lipstick; the project scope defines which lines and how many shades are included, in which channels the try-on appears and how the result connects to your catalogue — for instance an "add to cart" that already carries the shade the customer just tried.

Privacy: nobody keeps the photo

The customer's photo lives in memory only for the duration of the request and is forwarded once to the model. The gateway never writes photos to disk, database or logs — logs record size, shade, finish, latency and status. The render service does not keep the image either: only the PNG comes back to you, with Cache-Control: no-store. There is no facial recognition, no identification of people and no inference of protected traits.

Integration: start today via API or have it delivered as a project

Both paths use the same model. The difference is who builds the experience and which shades go in.