How This Color Analysis Works

The model, and where it comes from

Hue, value and chroma are not invented for this site. They are the three dimensions of Albert Munsell's color order system, published in the early twentieth century and still the standard vocabulary for describing color precisely. Hue is the temperature axis, value is lightness, chroma is saturation.

Applying them to human coloring is the core idea behind every serious color analysis system. Carole Jackson's Color Me Beautiful (1980) popularised the four season version using temperature alone. The twelve season systems that followed, largely descended from Sci\ART, added value and chroma and asked which of the three dominates a given person. That question is what this tool answers.

How the photo path measures you

1. 01

   Find the face

   MediaPipe's face landmark model runs in your browser as WebAssembly and returns 478 points, including the iris. No image data leaves the page.

2. 02

   Sample regions

   Both cheeks and the forehead for skin, a band above the hairline for hair, the iris for eye colour, and the corners of the eyes for the sclera. Each region uses the median of a patch rather than a single pixel, so a specular highlight or a blemish cannot move the reading.

3. 03

   Correct the light

   The whites of your eyes are close to neutral in most people, so they work as an approximate white point. Scaling the channels until the sclera reads grey removes most of the colour cast from your lighting. The size of that correction is recorded, because a large correction means an unreliable measurement.

4. 04

   Convert to CIELAB

   sRGB is not perceptually meaningful, so samples are converted to CIELAB where lightness is separated from the two colour axes and distances correspond to how different two colours actually look.

5. 05

   Derive the three axes

   Hue comes from the angle of your skin in the a*/b* plane combined with a warm or cool read of your hair. Value comes from the combined lightness of skin and hair. Chroma comes from the contrast between skin, hair and iris together with the saturation of your skin.

6. 06

   Send three numbers

   Those three values, plus a capture quality score, are all that is transmitted. The photo stays on your device.

How the season is chosen

The axis furthest from neutral is your dominant dimension and it names the season. A dominant value axis makes you a Light or a Dark season, a dominant chroma axis a Soft or a Bright one, and a dominant hue axis one of the four True seasons. The hue reading then picks the side: warm or cool.

Confidence combines how far the dominant axis sits from neutral with how far ahead it is of the second axis. A person who is strongly light and only mildly anything else gets a high score. A person whose three axes are nearly equal gets a low one and is shown both candidate seasons, because in that situation the honest answer is that they are on a border.

What we do not claim

Every competitor in this space advertises an accuracy figure, usually between 94 and 97 percent. There is no published benchmark for color season classification, no agreed ground truth, and professional analysts disagree with each other on borderline cases, so a number like that cannot mean anything. We do not publish one.

It is worth separating two claims that usually get bundled together. That colour near a face changes how the face is judged is well supported: given control of the colour of 51 faces along the CIELAB axes, people reliably shifted them toward a healthier appearance, raising redness by 1.62 units, yellowness by 5.25 and lightness by 1.21, all at p < 0.001 (Stephen and colleagues, 2009). A 2026 study in Coloration Technology showed the same thing with clothes rather than pixels: across 22 plain sweatshirts, the colour of the garment measurably changed how fair the wearer's skin appeared. That is the draping effect, measured.

What is not established is the sorting of people into twelve named boxes. No peer-reviewed study validates the twelve season taxonomy or reports how often two trained analysts agree. Treat the framework as a useful shorthand built on a real effect, which is what it is, rather than as a measurement of you.

What we publish instead is a per result confidence score and the axis readings behind it, so you can see whether your own result was clear or marginal. Lighting is the largest single source of error, dyed hair the second, and makeup the third. When confidence comes back low, the twelve question quiz is a genuinely independent second opinion, because it fails in different places than the camera does.

Where the color data comes from

The palettes are built from a curated dataset covering all twelve seasons: the defining dimensions, skin, hair and eye profiles, sister season relationships, best metals, makeup direction and colors to avoid. It is cross checked against theconceptwardrobe (Sci\ART derived), 12 Blueprints, Color Me Beautiful and Munsell's own system.

The hex codes are expert representative approximations, and this is worth being blunt about: no major analysis system publishes official hex values, because their draping swatches are proprietary physical fabrics and they differ between systems. Treat the hex codes here as an accurate visual guide to each color family's temperature, depth and clarity, not as a certified swatch you could send to a dyehouse.

Every color in the dataset is checked mechanically before it ships: each one must sit inside its season's measured lightness and chroma envelope, no two colors in a season may be so close that they render as the same chip, and the balance of warm and cool colors in a palette has to match the undertone the season claims. A color that fails any of those checks fails the build rather than reaching a page.

Inclusivity is a technical requirement, not a statement

Undertone is independent of skin depth. People of every skin depth can be warm, cool or neutral, and all twelve seasons occur across the full range of skin tones. Any tool or article that treats deeper skin as automatically warm is describing a population average and applying it to an individual, which is simply an error.

Two consequences for how this tool is built. The skin sample is taken from the median of several patches across cheeks and forehead rather than a single point, which matters more at deeper skin tones where specular highlights carry a larger share of the signal. And the value axis is derived from skin and hair together rather than skin alone, so depth of skin does not by itself push someone toward the Dark seasons.

The vein test and the sun reaction test are both presented on this site only to explain why they do not work. Both partly measure melanin rather than undertone, and both are least reliable for deeper skin, which is exactly where most published advice is already weakest.

Privacy, concretely

There is no upload endpoint on this server. Not one that is disabled, not one behind a flag: the route does not exist. Face detection and colour sampling happen in your browser, and the only request made after analysis carries three numbers and a quality score.

You can check this yourself in about thirty seconds. Open your browser's developer tools, go to the network tab, run the analysis, and look at what was sent. This is the kind of claim that should be verifiable rather than trusted.