Method, honestly
How it works
ColorWhim looks at the colors of your skin, eyes, lips and hair in a photo, estimates three qualities, and suggests which seasonal palettes might be worth trying. Then it hands the decision back to you with virtual drapes.
For fun and exploration. ColorWhim is not a professional color analysis and doesn’t claim to be accurate. A camera only sees what the light lets it see.
1. Everything happens on your device
When you open the tool, your browser downloads two small machine-learning models (about 20 MB) and a WebAssembly runtime from colorwhim.com itself. No third-party servers are involved. Your photo is processed in your browser’s memory and is never uploaded, stored or shared. Close the tab and it’s gone. See our privacy policy.
2. Finding the face and the regions to sample
A face-landmark model (MediaPipe Face Landmarker, 478 points) finds the outline of your face, eyes, irises and lips. A second model segments the image into hair, face skin, body skin, clothes and background. From these, ColorWhim draws small sampling regions:
- Cheeks and forehead for skin (only pixels the segmenter agrees are skin);
- Iris (a ring, avoiding the pupil and reflections) for eye color;
- Lips (the outer lip, not the mouth opening);
- Hair above the forehead, or your brows if there isn’t enough visible hair;
- Eye whites, used only as an optional reference.
Within each region we trim the darkest and brightest pixels (shadows, shine, lashes) and take the median, which is robust to small mistakes.
3. Correcting the light
Your camera’s idea of “white” is a guess. If you hold a sheet of plain white paper or a grey card in the frame, ColorWhim tries to find it automatically (or you can tap it) and uses it to neutralize the light’s color. Without one, it estimates the light from the rest of the scene (a “gray-world” estimate), which works less well with colorful walls or clothes.
In our own stress test (26 openly licensed portraits, each re-rendered under nine simulated lighting and quality changes), the suggested season family flipped in roughly 1 in 4 cases with the whole-scene estimate, versus roughly 1 in 7 with a white or grey reference. That’s a measure of stability, not accuracy, but it shows why the white paper helps.
4. From colors to three axes
Samples are converted to CIELAB, a color space designed around human perception. From the skin we compute the hue angle (more yellow means warmer, more pink-red means cooler) and chroma (how saturated). We also compute lightness for skin, hair and eyes and the contrast between them. These are combined into three scores between −1 and +1:
| Axis | What feeds it |
|---|---|
| Warm ↔ cool | Skin hue angle (mostly) and lip hue |
| Light ↔ deep | Lightness of skin, hair and eyes, weighted |
| Clear ↔ soft | Skin chroma and skin–hair–eye contrast |
Each score is a smooth curve around a pivot point we chose by hand. Those pivots are informed guesses: they have not been calibrated against professionally draped people.
5. Suggesting seasons
Each of the 12 seasons has a reference position on the three axes. ColorWhim measures how far your scores are from each one and turns the distances into relative “leanings” that add up to 100%. We show the top two. A high leaning means your photo sits clearly inside one neighborhood according to our rules; it is not a probability that the answer is correct. We also show the season that sits farthest away as colors to approach carefully.
6. Quality checks
Before you look at results, ColorWhim flags things that commonly distort them: a face that’s too small in the frame, blown-out highlights, a very dark photo, uneven light between the two sides of your face, a very colorful background, an over-exposed white paper, unusually smooth skin texture (a hint of beauty filters) and editing-app tags in the file. These are heuristics; they can miss problems or warn unnecessarily.
7. Virtual drapes: the part that matters most
Professional analysts hold fabric under your face and watch how your skin responds. ColorWhim imitates this by drawing a colored cape below your chin, shaded to follow your photo’s light. Flip through all 12 seasons and compare side by side. Your eyes are a better judge than any score, and the drapes let you use them.
Known limits
- Lighting: warm lamps, mixed light, shade and direct sun all shift colors. Soft daylight from a window works best.
- Filters and editing: beauty modes, portrait filters and heavy compression change skin color and texture. Use your phone’s default camera with filters off.
- Makeup, tans and hair dye change what the camera sees. Bare skin and natural hair give the most “you” result, though dyed hair is part of your look too.
- Screens differ: your display’s calibration affects how drapes and swatches look.
- Color management: photos are treated as standard sRGB; wide-gamut (Display P3) photos are read without conversion, which can slightly change colors.
- No ground truth: the rules haven’t been validated against professional analyses. For context, published research on automatic seasonal classification reports roughly 50–70% agreement for the four families on a benchmark dataset, so even well-trained systems disagree with experts often.
FAQ
Is ColorWhim accurate?
We don’t claim accuracy. Nobody has professionally draped the people in our test photos, so we cannot measure how often ColorWhim agrees with a trained color analyst. Treat the result as a starting point for your own comparison.
Is my photo uploaded?
No. The models and the code run inside your browser. Your photo is decoded, analyzed and drawn on your device and is never sent to our servers or anyone else.
Why do I get different results from different photos?
Lighting changes the colors a camera records. Warm lamps, colored walls, shade, auto white balance and beauty filters can all move the estimate, sometimes across a season border. Photos in soft daylight, with a white paper in the frame, give the most consistent results.
Does it work for every skin tone?
The method measures color relationships (hue, lightness, chroma and contrast) rather than assigning seasons by skin depth, and any skin tone can land in any season. However, the face models and our rules have not been validated on a large, diverse, professionally labeled set, so results may be less reliable for some people and some lighting setups.