Inside Visage AI

Due to the increase in inbound tourists, the frontlines of beauty retail are facing a huge opportunity. However, at the same time, the repeatedly occurring problem of the "language barrier" is becoming a burden for staff.

"A base suitable for dry skin," "Highly pigmented lipstick that brightens the skin of a cool tone winter" — the reality is that such subtle nuances and beauty jargon are difficult to convey accurately with ordinary translation apps.

Visage AI does not fill the language barrier with "translation," but rather "converts" it into a form that can be used on-site by utilizing non-verbal biometric data.

By instantly calculating the facial bone structure, skin uniformity, and contrast of features, it connects to an objective proposal flow that store staff can intuitively understand.

In this article, we reveal the backstage and design philosophy of the analysis engine that makes Visage AI a B2B tool robust enough for business use.

1. The B2B Workflow: 3 Steps to Beautify Customer Service

The Visage AI experience is completed on a store iPad (or the customer's smartphone). It is an elegant workflow designed to shorten service time and reduce variations in proposal quality among staff.

1. Scan & Validate

For more accurate analysis, we recommend live scanning on the spot in store mode. Loading images from the gallery is also possible, but only images that pass the strict "Quality Control Gate" described below will proceed to analysis.

2. Translate to Product

When analysis is complete, specific proposals such as "Looking for: Lip liner to adjust contour / High moisturizing concealer" are displayed on the screen. The customer simply shows this to the staff, and accurate product guidance begins.

3. Secure Wipe

For store use (Guest Mode), the session is discarded with a single tap at the end of the customer service. It implements a flow intended to delete related data at once, including the device's cache and storage area, designed with privacy protection in mind.

2. Under the Hood: Uncompromising "Biometric Analysis Engine"

In the background of Visage AI, we use on-device face contour detection (native pipelines like ML Kit) and asynchronous processing (Isolates) to perform fast and secure analysis.

Quality Control Gate

The biggest challenge in image analysis is the quality of the input image. Visage AI places a strict Quality Control (QC) gate before running the analysis engine.

It automatically detects images with insufficient illumination or blur, and blocks analysis. "Prioritizing not outputting a result over outputting an incorrect result" — this philosophy is designed to enhance operational reliability in the field.

Vermilion Border Index

Signs of aging and dryness tend to appear on the Vermilion Border (the outline of the lips).

Visage AI extracts the outer area of the lips from facial landmark detection and calculates the luminance gradient (sharpness of the edge) from the inside of the lips toward the outer skin. If the border is judged to be ambiguous, the system adds items like "lip liner" or "lip serum" to the recommendation list, supporting evidence-based proposals.

Periorbital Topography

To evaluate "dark circles" and "dullness around the eyes," we analyze using the CIELAB color space (L*a*b*) which is close to human vision, rather than simple RGB values.

It compares the standard color of the cheek (base skin tone) with the color difference (ΔE) and lightness difference (ΔL*) in a specific area under the eyes. By doing this, it evaluates the three-dimensional shadow of the face (such as eye bags) separately from dullness caused by coloration (Pigmentation), leading to optimal concealer or eye cream proposals.

Future Care Engine (Labs)

Beyond current skin analysis, Visage AI introduces a preventive logic to identify long-term care priorities. By weighting current biometric scores against statistical vulnerability, we surface the areas that require the most attention for sustained skin health.

Native OS Decoding (HEIF Support)

To ensure high reliability on the shop floor, Visage AI utilizes native OS-level decoders for image processing. By supporting HEIF/HEIC formats taken directly with iPad cameras through robust system pipelines, we have significantly improved compatibility and reduced failures during the analysis flow, ensuring a seamless experience for both staff and customers.

3. What Visage AI Does Not Do

To operate safely in a retail environment as a B2B tool, Visage AI strictly limits its scope of use.

• Does not perform medical diagnosis
• Does not estimate psychology or personality
• Is not intended for personal identification purposes
• Does not analyze images that fail to meet quality conditions (e.g., poor lighting)

4. Visage Labs: Latest Innovations

The Visage AI engine is constantly evolving through our R&D pipeline. Following the release of v3.2, these key innovations have been integrated into our production environment:

• Future Care Navigator (Labs): Focuses on long-term skin health by prioritizing care areas based on biometric scores and statistical weighting.
  *This feature provides preventive guidance and is not intended for medical diagnosis or clinical treatment.
• Facial Geometry Profiler: Based on geometric features such as the ratio of the face's width and height, and the balance of the contour, it proposes the optimal placement of shading and highlighting.
  *This feature is a geometric auxiliary tool intended for makeup style proposals and is not intended to estimate an individual's psychology or personality.

To achieve both transparency of analysis and the beauty of the customer service flow. That is our design philosophy.

The Visage AI analysis engine will cross the "language barrier" of stores and provide a better experience for customers.