How we measure your colors
No quiz and no guessing from a description. We read the color of your skin, hair and eyes from the photo itself, as numbers, and we tell you where that goes wrong.
The short version
flatterlab measures color from a photo in four steps. First, a face model (MediaPipe Face Landmarker) finds 478 points on the face, and a second model marks which pixels are skin, hair and clothes. Second, if a sheet of white paper is in the photo, flatterlab reads the paper and scales the photo until the paper is neutral white at a fixed brightness. With no paper it keeps the phone's own white balance and lowers its confidence by one step. Third, it samples skin from four spots (both cheeks, forehead and chin), the iris of both eyes and the hair, drops the brightest and darkest pixels, and takes the middle color in CIELAB. Fourth, it turns those numbers into four scores, which are undertone, depth, clarity and contrast, and picks the nearest of 12 seasons.
How we measure, and where it fails.
-
Find the face.
A face model that runs on your device marks your skin, hair, eyes and lips.
-
Correct the light.
The white paper shows the tint of your light. We take that tint out.
-
Read the colors.
We sample each area and record lightness, warmth and strength as numbers.
-
Compare with all 12 seasons.
Four scales decide it: undertone, depth, clarity and contrast. The free tool names the family. The paid report names the nearest season and the runner-up.
It gets these wrong
Bad light. Lamp light, shade and flash shift every number. The photo check stops the worst cases, not all of them.
Heavy makeup. Foundation and bronzer hide the skin we need. Come with a bare face.
Dyed hair. We measure the hair in the photo. Dyed hair shifts the depth and contrast scores, so the reading fits the hair you have now.
Screens. Phones show color brighter than fabric does. Turn off night mode and check your palette in daylight.
Same photo, same answer, every time. A different photo in different light can move you to the season next door. That is why the paid report shows the runner-up.
-
The free tool
Your photo stays in your browser.
-
The paid report
Measured on our own computer, never sent to an outside AI service.
-
After delivery
Photos for paid orders are deleted 30 days after delivery.
The four scores
| Score | What we measure | The two ends |
|---|---|---|
| Undertone | The hue angle of your skin color | Cool (pink) to warm (golden) |
| Depth | Lightness of skin, hair and eyes, weighted 50, 30 and 20 percent | Deep to light |
| Clarity | Color strength of skin and eyes, weighted 60 and 40 percent | Soft to clear |
| Contrast | The gap in lightness between hair and skin | Low to high |
Each of the 12 seasons has a target on all four scores. Your season is the nearest target. When the nearest two are almost the same distance away, we say so and call it a close call.
What the photo check refuses
A wrong photo gives a wrong season, so the tool stops before it measures.
| We refuse a photo when | Why |
|---|---|
| Face under 150 pixels wide | Too few skin pixels to take a stable reading. |
| Head turned or tilted more than 15 degrees | One cheek falls into shadow and the two sides stop agreeing. |
| Colored light | A lamp or a sunset pushes skin past the range real skin has in daylight. |
| A filter or beauty mode | Retouched skin reads pinker than skin can be. |
| Black and white | There is no color left to measure. |
| More than one face, or none | We must know whose colors we read. |
Accuracy and limits
We do not print an accuracy figure, because we do not have one yet. The cut-off numbers above are our first settings. We set them on a small set of test photos, and they will move as we check more real ones.
Three things to know before you trust a result:
- The free tool reads one photo. A paid report reads up to three. With two or more it checks that they agree on the season family, and if they do not, we ask you for a new photo. With one photo that check cannot run, so send two.
- Without white paper we cannot remove the tint of your light. The result is then a first read.
- A person looks at every paid report before it is sent. Nobody looks at a free result.
What is made by AI on this site
The text and the drawings on this site were made with the help of AI tools. The sample report uses an AI-generated test photo, not a customer. Your result is different: code measures it from your pixels. No language model writes your season, and your photo is never sent to an outside AI service.