ChatGPT color analysis prompt you can copy and improve

I'm attaching photos of myself, taken in daylight with no filter. I want a careful personal color estimate, not a verdict.

Before you answer, tell me what you can and can't judge from these photos: lighting, white balance, makeup, hair dye, image quality. If something would change the answer, say so.

Then rate each of these from -2 to +2, and give the evidence you can see for each:
- Temperature: cool (-2) to warm (+2)
- Depth: light (-2) to deep (+2)
- Clarity: soft (-2) to bright (+2)

Base depth on my overall coloring, including my hair and eyes. Don't treat my skin depth or ethnicity as an answer.

Then give me:
1. The season in the 12-season system that fits best, and a runner-up.
2. How sure you are, in words, and what would change your mind.
3. Six colors to test near my face and three likely to be bad, in plain words, each with a reason.
4. One test I can do with real clothes by a window to separate your top two seasons.

If the photos disagree with each other, tell me instead of averaging them. Don't flatter me, and don't treat the season as a fact about me.

Why the prompt is written this way

A typical color analysis prompt asks “what season am I?”, and you get a season, stated with confidence. The prompt above is built to make the weak parts show. Each section has a job.

• “Tell me what you can and can’t judge” comes first, so that the model commits to its doubts before it commits to an answer.
• The three ratings split the question into temperature, depth and clarity. A vague season name is hard to check, while “clarity -1, because the skin looks fine against muted tones” is something you can look at.
• “Base depth on my overall coloring” heads off the old mistake of equating dark skin with a dark season.
• A runner-up and what would change its mind make it hold a second idea. Many people sit between two seasons anyway.
• A test with real clothes gives you a way to overrule it, which is the whole point.

We can’t promise that this prompt gets you a right answer. No wording can fix a photo with bad white balance. What it should do is give you an answer that’s easier to check.

What photos should I upload?

The photos matter more than the prompt. Here’s what we’d send:

1. One close-up facing a window, out of direct sun, with no flash, makeup or filter. Pull back any dyed hair.
2. One taken the same way while holding plain white printer paper next to your face. The paper gives the model something known to compare against. Avoid letting it go blown-out white.
3. One taken outdoors in shade, or at another time of day. If the answers change between this and the first, you’ve learned something about the method.

Wear white or plain grey, and stay away from colored walls. Turn off your phone’s warm screen mode and any portrait or beauty mode. A popular walkthrough suggests much the same setup: 2 to 3 selfies in diffused daylight, in a neutral top or with bare shoulders. A face photo is personal data, so check your settings first. The ChatGPT page has what OpenAI says about that.

If you’d rather not upload a photo

This is much rougher, as StyleCard also points out: you’re translating your face into words and then the model is interpreting your words. It’s fine as a first pass, and the prompt tells it to be honest about that.

I don't want to upload a photo. Here's my natural coloring, as best I can describe it:

- Skin: [how light or deep; does it burn or tan; any redness or sallowness]
- Hair: [natural color now, and as a child; is it dyed]
- Eyes: [color, and how much the iris stands out from the white]
- Colors people compliment me in: [list]
- Colors that make me look tired or sick: [list]

From this alone, give me a rough estimate. Tell me plainly how little you can tell from a description.

Rate temperature (cool -2 to warm +2), depth (light -2 to deep +2) and clarity (soft -2 to bright +2). Then name the two 12-season types you'd test first, and for each, three colors I can hold against my face by a window, with what I should look for.

Follow-up prompts, in the order we’d use them

The first answer isn’t the useful one. These push on it, one step at a time, and you can stop whenever you have what you came for.

Now make the strongest case for your runner-up season instead. What in my photos supports it? What did you weigh too lightly the first time? Finally, tell me whether your first answer still stands, and how close the call is.

In this photo I'm holding a sheet of plain white printer paper in the same light. Does the paper look neutral in the picture, or does it lean warm, cool, green or pink? If it leans, tell me which way and roughly how much. Then redo your temperature rating, allowing for that cast, and say whether your season changes.

Here are two photos of me in the same light and pose, one in a [color A] top and one in a [color B] top. Look only at my face: evenness of the skin, shadows under the eyes and around the mouth, and how clear my eyes look. Which photo looks better on those points, and how sure are you? If you can't tell, or if the difference could just be the photos, say that instead of choosing.

I'm trying to choose between [season 1] and [season 2]. Give me three pairs of colors, one from each season, close in lightness and strength, that I can find in ordinary clothes. For each pair, tell me what I should see on my skin if season 1 is right and what I should see if season 2 is right. I'll do this by a window and report back.

Assume [season] is right for now, and that I may change my mind. Give me: 3 neutrals to build on, 6 colors to wear near my face, and 3 to keep away from it. Then a 12-piece capsule wardrobe using only these colors for [my climate and lifestyle], and exact words I can type into a store's search for each piece. Name colors in plain words, and don't invent product names or brand shades.

Prompt 4 is the one we’d never skip. After that you can bring in lipstick families, glasses or hair. If you want more than a chat can give, the season test reads your skin, eyes and hair from your camera, corrects the light with white paper, and shows your season, palette and colors beside your photo.

Should I use hex codes?

You’ll see this hack everywhere: sample your skin, hair and eyes with a color picker and paste the hex codes. It feels scientific, and we’d be careful with it. As StyleCard puts it, “A hex code captures the color of one pixel under one lighting condition”. A skin pixel also carries shadow, redness, compression and whatever the camera decided about white. One person who tried it with a known result got the right softness and the wrong undertone, and Perfect Corp warns that results may be inconsistent.

If you want to do it anyway, here’s the same idea with one thing added: a sample of something that should be white. It gives the model a way to estimate the camera’s error instead of taking the skin number at face value.

I sampled these colors from one daylight photo with a color picker. They carry the camera's white balance, so treat them as approximate.

White paper in the same light: #______
Skin, cheek: #______
Skin, jaw: #______
Hair: #______
Iris: #______
White of the eye: #______

First, tell me how far the paper is from neutral and in which direction. Then estimate how each sample would shift if the paper were truly neutral. Show the corrected values. Then rate temperature, depth and clarity from the corrected values, and name two 12-season types to test. Say clearly what a handful of pixels can't tell you.

Can it make images of me in my colors?

Often, yes, with an image model on your plan. It’s fun, and it can help you imagine a color. Don’t use it as evidence, though. The model may change your skin, your features or the light as it goes, and it’s painting a picture, with no real fabric reflecting real light on you. If two images look different, you can’t tell whether the color did that or the model did.

Edit this photo of me. Keep my face, skin tone, hair, expression, lighting and background exactly as they are. Change only my top to a plain [color name] with no pattern. Don't retouch, smooth or lighten my skin. If you can't keep my skin unchanged, tell me.

How to read the answer

• Check the ratings against what you can see. If it calls your eyes high contrast when they aren’t, the rest is built on sand.
• Be wary of flattery. Being told you’re a rare, striking type is pleasant and tells you nothing.
• Ask for colors, not just a label. Names vary between systems, as the 16-season page shows, and a “Cool Summer” can mean two things. Get the colors in plain words.
• Repeat it in a new chat. The tally on the ChatGPT page helps you see how often it agrees with itself.
• Trust cloth over chat. When a real top by a window disagrees with the model, the top wins.

If you want the same questions answered by something built for them, try the free scan and compare.

Related guides and tools

• ToolWhat season am I?Three axes narrow down your season, then color pairs beside your photo settle it.
• GuideChatGPT color analysisWhat a chatbot can infer from a photo, why answers shift and how to check them.