ChatGPT color analysis: why it guesses, and how to get a measured answer
Asking ChatGPT to name your color season from a selfie is a reasonable first move. The catch is that a language model describes your photo; it cannot measure it, show you the result, or correct for your camera. Those three are the whole job.
By Ruslan ShchuchkinML engineer and data scientistUpdated
Three things a chat cannot do
- 01
Show the color on your face
A chat returns color names. Here your face is cut out and set on each color, and in the full report the color is reflected up onto your jaw and cheeks, the way real fabric under the chin lights the face. That reflection is the actual test of whether a color suits you, and no chat can run it.
- 02
Measure instead of describe
A chat writes the most plausible answer about your photo. Here 468 facial landmarks sample your cheeks, forehead, irises and hair, and the numbers decide the season, so the same photo always gives the same result. It runs in your browser: the photo is never uploaded and never stored.
- 03
Correct for the camera
A camera’s white balance shifts warm against cool before any model sees the image, which is why a chat names your undertone confidently and often wrongly. Here each photo is calibrated against the whites of your eyes, lighting is capped in how far it can move the reading, and photos in different light are combined so each cast cancels.

Deep teal is in her season. With the reflection on, it lights her face from below the way a good drape does.
This is the demo model, a True Autumn, not you. The same reflection runs on your own face in the full report’s single-color view; the free report drapes your face without the cast.
ChatGPT versus a measured analysis
Two rows go to the chat: it is free and it will talk you through the idea. The rest are the difference between describing a face and measuring it.
| ChatGPT | Here (free) | |
|---|---|---|
| Your photo | Uploaded to a server and processed there. | Measured in your own browser, never uploaded. There is no upload endpoint on this site. |
| What it does with it | Describes it, in words. | Measures it: 468 landmarks, converted to CIELAB, a perceptually even color space. |
| Same photo, asked twice | Often two different seasons. There is no fixed scale underneath. | The same result every time. It is a measurement. |
| Undertone | Read off the raw photo, which the camera has already shifted. | Read off a calibrated photo: white-balanced against the whites of your eyes, lighting capped, photos in different light combined. |
| Seeing the color on you | A list of names, and hex codes it may have invented. | Your face on all twelve palettes, free. In the full report, one color at a time, with the color reflected onto your face. |
| How sure it is | Usually rounds to one season. | All twelve scored; the runner-up and the exact gap are always shown. |
| Cost | Free, inside a chat. | Free for your season, confidence, runner-up and starter palette. $9 for the full report, 30-day refund. |
Your photo
- ChatGPT
- Uploaded to a server and processed there.
- Here (free)
- Measured in your own browser, never uploaded. There is no upload endpoint on this site.
What it does with it
- ChatGPT
- Describes it, in words.
- Here (free)
- Measures it: 468 landmarks, converted to CIELAB, a perceptually even color space.
Same photo, asked twice
- ChatGPT
- Often two different seasons. There is no fixed scale underneath.
- Here (free)
- The same result every time. It is a measurement.
Undertone
- ChatGPT
- Read off the raw photo, which the camera has already shifted.
- Here (free)
- Read off a calibrated photo: white-balanced against the whites of your eyes, lighting capped, photos in different light combined.
Seeing the color on you
- ChatGPT
- A list of names, and hex codes it may have invented.
- Here (free)
- Your face on all twelve palettes, free. In the full report, one color at a time, with the color reflected onto your face.
How sure it is
- ChatGPT
- Usually rounds to one season.
- Here (free)
- All twelve scored; the runner-up and the exact gap are always shown.
Cost
- ChatGPT
- Free, inside a chat.
- Here (free)
- Free for your season, confidence, runner-up and starter palette. $9 for the full report, 30-day refund.
Run it against whatever ChatGPT told you
Keep the season the chat gave you, take the free analysis here, about three minutes, and compare.
- If they agree, you have a measured confirmation with a confidence score behind it.
- If they disagree, it is almost always on undertone, the axis the chat read off a raw photo and this analysis calibrates for.
- Either way you see it on your own face instead of imagining a list of color names.
When ChatGPT is the better tool
For the conversation. What the twelve seasons are, why undertone is treated separately, how to adapt a color you love that is not in your season: a chat is excellent at all of that. Only the measurement, the actual answer, is the part it cannot give you reliably.
You can read a complete sample report first, or see how the method works and where its limits are.
Frequently asked
Can ChatGPT do color analysis?
It can attempt it, and the answer reads well, but it is describing your photo, not measuring it. There is no fixed color space underneath, so it cannot score you consistently against the twelve seasons. Ask twice and the season often changes.
Why does ChatGPT give me a different season each time?
Because nothing fixed sits under the answer. A measurement converts the photo to numbers and compares them to each season, so the same input lands in the same place. A language model generates text that fits the prompt, so rewording it, or simply asking again, moves the result.
Is ChatGPT accurate for finding my color season?
Best on obvious cases, weakest on undertone, which is the axis that matters most. A camera’s white balance shifts warm against cool before anything reads the image, so any tool reading undertone off a raw photo is guessing. This site calibrates the photo first, and where two seasons come out close it shows the runner-up and the gap rather than false certainty.
What does this tool do that ChatGPT does not?
Three things. It measures the photo with 468 landmarks instead of describing it. It corrects the photo for the camera before reading undertone. And it shows you the result on your own face, with the color reflected onto it in the full report, rather than handing you a list of color names.
Should I use ChatGPT or a dedicated color analysis tool?
Use the chat to understand the idea: what the seasons are, why undertone matters, what to do with a palette. Use a measured tool for the answer itself, because that is the part a language model has to guess. Running this one against whatever ChatGPT told you takes about three minutes and is free.
Is it safe to upload my face to ChatGPT?
A photo sent to any chat assistant is uploaded to a server to be processed. Here it is measured in your browser and never leaves your device; there is no upload endpoint on this site, which you can verify in your browser’s network tab.