How Our AI Color Analysis Works, and How We Make It More Accurate
The ShopYourHue color test uses AI to read your coloring from a photo of your face, and then it checks that reading against 7 quick questions before it names your season. A photo on its own is easy to fool, because light, makeup and even the wall behind you change how your skin looks on camera, so we built the color test to catch those mistakes instead of passing them on to you.
AI can tell a lot about your coloring from a photo, but not everything, and the parts it gets wrong are the ones that decide your season. This post explains what the AI reads, where a photo lets it down and what we do about it. ShopYourHue never saves your photo, and my post on free seasonal color analysis online explains what happens to it.
What the AI looks at
When you take a photo, an AI model looks at it and describes your coloring. It reads how light or deep you are, your undertone, how bright or soft your coloring is, your hair color, your eye color, and how much your hair stands out from your skin.
Some of these are much easier for a camera than others. Light or deep comes through in almost any photo, because dark hair looks dark under a lamp and in daylight. Warm or cool and bright or soft are harder, because they are exactly what the light in your room changes.
The AI only describes what it sees. It doesn't pick your season. Your season is worked out by our own scoring, which runs in your browser and weighs the photo together with your answers.
Why a selfie on its own gets it wrong
A camera doesn't see your skin the way a mirror in daylight does. A ceiling lamp makes everyone look warmer, and a cloudy window can make you look cooler. Foundation, bronzer and beauty filters cover the very skin the AI is trying to read, and a bright red top or a yellow wall bounces its color up onto your face. Phone cameras also change colors on their own before the AI ever sees the picture, warming up a dim room or smoothing your skin.
A test that reads one selfie and gives one answer takes all of those mistakes as the truth. Two photos of the same person in two rooms can come out as two different seasons, and the person has no way of knowing which one to believe.
What we do differently
Each step of the color test is there to stop one of those mistakes.
- A short checklist before the photo. We ask about makeup, filters, light, what is behind you and what you are wearing. Whatever you can't do, we trust that part of the photo less, and a "Photo confidence" bar shows you how much we can trust it before you take it.
- White paper by your face. If you hold a plain sheet of white paper beside your cheek, we know what white looks like in your light, so we can correct the color of the whole photo. This is the single most helpful thing you can do.
- Your answers count too. After the photo come 7 quick questions, such as whether gold or silver looks better on you in a mirror. Some things are easier for you to judge in person than for a camera to judge from a photo, so your answers are weighed together with the photo.
- You confirm what the photo saw. Your hair color, eye color and contrast are already picked from your photo, and you can keep our pick or change it. If your hair is dyed or you wear colored contacts, we ask for your natural color instead.
- Two seasons when it's close. Some people sit between two neighboring seasons. When your result is close, the color test shows both palettes instead of guessing, and you can shop them together.
This is what makes it more accurate than a photo on its own. No single photo has to be perfect, because the color test knows how much to trust it and has your answers to check it against.
How we test it
We test the color test on photos of people whose seasons are already known, some of them clients of trained color analysts. When it gets someone wrong, we look for the reason, whether it was the light, the hair or the undertone, and we change the test where we can.
Something interesting we found is that the AI makes the same mistake on the same photo every time. Sending a photo again doesn't fix it, and the newer and bigger AI models we tried didn't do any better. What changed the result was the photo itself and the answers that go with it, which is why the color test puts so much into the light, the white paper and the 7 questions.
We also found that dyed or sunlit hair could outweigh everything else in a photo, so the hair and eye color the AI sees now count for less than the ones you confirm yourself.
The color test is a free guide, not a professional draping. It is a good way to find your season, or to narrow it down to two before you decide whether a draping is worth it for you.
The takeaway
AI is good at describing a face, but a photo can only show what the light lets it show. Our color test treats the photo as a strong first reading, corrects it where it can, and checks it against what you see in the mirror, so the season it gives you rests on more than one photo.
When you are ready, take the color test by a window with a sheet of white paper. My post on free seasonal color analysis online goes through what you need and every step.