Can ChatGPT Do Your Color Analysis? Yes, With Limits
Yes, ChatGPT can do a color analysis, with limits. Upload a clear photo and it will usually name a color season, explain its reasoning, and suggest a palette. What it cannot do is see your true coloring: it reads whatever the camera and the light recorded, it can give a different season when you ask again, and chatbots lean toward agreeing with whatever you suggest.
Not sure which season is yours? One photo reads your undertone, depth and contrast, and names it. About a minute, $9.
Find my seasonThis page explains what a general chatbot is doing when it types you, how to get the most reliable read out of it if you want to try, and how that compares with a purpose-built analysis and an in-person draping.
A note on where we stand: Colorito sells an AI color analysis, so we have a side in this. The documented limits below come from published sources, listed at the end; the rest is our reading of how these tools behave. We have not benchmarked ChatGPT, and we do not quote an accuracy figure for it or for ourselves.
Key takeaways
- •Yes, with limits. ChatGPT and other chatbots with image input can estimate a color season from a photo and explain why.
- •The photo is the weak point. A phone's white balance, indoor bulbs, and filters shift skin toward warm or cool before any AI sees it.
- •Answers can change between chats, and AI assistants are known to tilt toward the user's stated view. Ask three times in temporary chats and compare.
- •A good prompt helps: daylight, no makeup, no filter, a white reference in frame, and an instruction to identify the lighting before judging undertone.
- •A purpose-built analysis uses the same kind of model with a fixed procedure around it: lighting correction, calibration questions, and repeated independent reads.
What ChatGPT Is Doing When It Types You
ChatGPT is a general-purpose model that accepts images. When you upload a selfie and ask for your season, it describes what it sees (hair, skin, eyes, the contrast between them) and maps that description onto the twelve-season system it learned from text about color analysis. It is doing a version of what a human analyst does from a photo.
It is not measuring anything. There is no colorimeter and no drape, only pixels. OpenAI's own documentation for its vision models says they "can make mistakes," may produce incorrect descriptions, and may resize images before analyzing them. None of that makes the exercise worthless. It means the answer is a reading of a photograph, and it is only as good as the photograph and the instructions.
Where a ChatGPT Color Analysis Goes Wrong
Five things work against a chatbot read. The first two apply to every photo-based analysis, ours included. The last three are what a fixed procedure is built to reduce.
- •Lighting and white balance. A camera adjusts the overall color of an image so that neutral things look neutral, and it guesses. Warm bulbs push everything golden; shade and overcast light push it cooler. Undertone, the first call in any color analysis, is a warm-versus-cool judgment, so the camera's error lands on the most important decision.
- •Filters, beauty modes, and makeup. Smoothing and brightening change the very skin the analysis is trying to read, and foundation covers the undertone with a different one.
- •Different answers on different tries. OpenAI's developer documentation describes its models as non-deterministic by default: outputs "may differ from request to request." One answer is one draw. Ask again in a new chat and a borderline face can come back as the neighboring season.
- •Agreement bias. Researchers studying AI assistants found they consistently tilt toward a user's stated views, a behavior called sycophancy. Tell a chatbot you think you are a Soft Autumn and you have changed the odds of hearing that you are right.
- •No fixed procedure. Nothing in an open chat fixes the method. Whether it corrects for the light, where on the face it reads undertone, and whether it lets your sweater color influence it all depend on how you asked.
A ChatGPT Color Analysis Prompt That Gets a Better Read
If you want to try it, set up the photo first. Face a window in indirect daylight with the room lights off. No makeup, no filter, no beauty mode. Keep your natural hair color visible, or say what it is. Hold a sheet of white printer paper beside your face so the image contains a true white. Take two or three photos, and add one of your inner wrist.
Then paste this: "Act as a 12-season color analyst. I am attaching unfiltered photos taken in indirect daylight with no makeup; the white paper in frame is true white. First, describe the lighting in each photo and any color cast you see, and correct for it. Then assess, in this order: skin undertone at the jawline and neck (warm, cool, neutral, or olive), value (light, medium, or deep) from hair, brows, and eyes together, and chroma (bright or soft). Ignore my clothing. Give your top season and a runner-up, the evidence for each, and what would change your mind. Do not ask what I think I am, and do not adjust your answer to please me."
Run it three times, in temporary chats so an earlier answer cannot carry over, without telling it your own guess. If all three agree, you have a decent working answer. If they split between two neighboring seasons, that pair is your honest answer for now, and each season page links a side-by-side comparison with its neighbors. If they split between a warm season and a cool one, the photo is the problem: retake it.
We have not benchmarked this prompt. It is built from the checks a careful analysis runs, most of them listed under how the read is made in our guide: identify the light first, read undertone at the jaw and neck, judge value from hair, brows, and eyes together.
What a Purpose-Built Analysis Does Differently
Colorito's analyst is built on the same class of large vision model as ChatGPT. The difference is not a secret model. It is the fixed procedure around it, listed below, which runs the same way on every photo.
It is still a photo, and it can be wrong, most often by reading cool coloring as warm when the light is warm. Poor light lowers the quality of any read, ours included, and for someone on the border between two seasons it can land on the neighbor.
- •The light is identified first. Golden-hour sun, warm bulbs, neutral daylight, overcast, or flash. A golden cast that covers skin, hair, and background alike is treated as the light, not the person, and discounted before anything else is read.
- •Three calibration answers. Before the read, you answer three quick checks: your wrist veins, which metal flatters you, and how your skin reacts to sun. They are weighed against the photo to correct for lighting bias.
- •Undertone is read at the jawline and neck, never from makeup.
- •Repeated independent reads. Every analysis is run at least twice, independently. If the reads disagree, more are run and the majority decides.
- •It never asks which season you think you are. The only things you tell it are the three calibration answers.
ChatGPT vs a Color Analysis App vs In-Person Draping
The three options differ less in intelligence than in control: how much of the lighting, the procedure, and the consistency is handled for you. In-person prices come from our index of US studio prices.
| Compare | General chatbot (ChatGPT) | Purpose-built AI analysis | In-person draping |
|---|---|---|---|
| Cost | Nothing beyond the plan you already have | $9 at Colorito | $250–$400 for a standard US session (median $250–$300) |
| Time | A few minutes, plus retakes | About a minute | 90 minutes to 2 hours |
| What reads your coloring | A general vision model, from your photo | A vision model following a fixed procedure, from your photo and three calibration answers | A trained analyst, watching fabric drapes against your face |
| Lighting | Whatever your photo has; not reliably corrected unless you ask | The analyst is told to identify the light and discount its cast before reading undertone | Controlled: natural or color-corrected light |
| Consistency | Can change from chat to chat | At least two independent reads; the majority decides | One analyst's judgment; analysts can disagree |
| What you leave with | A chat reply | Season, evidence, palette, and colors to avoid, saved to your account; a runner-up season is shown when the read is close | Your season and a physical swatch palette |
| Best for | A first hypothesis | One decided season, with the evidence behind it | The full experience, and the hardest borderline cases |
Which One Should You Use?
If you are curious and already use ChatGPT, try it. Follow the photo setup above, run it three times, and treat agreement as a hypothesis, not a verdict. It is a reasonable first look.
If the three runs disagree, or you want a palette you can act on without second-guessing it, use an analysis built for the job. That is the gap Colorito's analysis is designed for: one decided season, the observations behind the call, your 10-color palette, and the colors working against you, plus the nearest neighbor when the read is close, saved to your account. It costs $9 and takes about a minute.
On the photo path there is a free preview before you pay: your season family, and whether you run warm or cool.
If you love the ritual, or two photo-based reads leave you on a border, book the draping. Seeing fabric change your own face in a mirror settles borderline cases in a way no photo does.
And if all you want is the practical output, the page on what colors look good on me gets you most of the way without any AI at all. If only your undertone is in question, the skin undertone test takes five at-home checks.
Want a decided answer instead of three chats?
Colorito reads your undertone, depth, and contrast from one photo, corrects for the light, and names your season with the evidence behind it. $9, about a minute.
Find your season for $9Keep exploring
The full guide to color analysis
Undertone, depth, contrast: how the verdict works.
What colors look good on me?
The practical answer, no AI required
What a color analysis costs
39 studios across 8 US cities
Skin undertone test
Five at-home checks, warm to olive
The skin tone chart
21 named tones: seven depths by three undertones
Olive skin tone
The undertone the usual tests misread most
See a full example result
A real analysis, start to finish: exactly what $9 buys
Frequently asked questions
Can ChatGPT do my color analysis?
Yes, with limits. Given a clear, unfiltered photo in daylight, ChatGPT will name a season and explain why. The read depends heavily on the photo's lighting, it can change between chats, and assistants tend to agree with a season you suggest. Run it three times in temporary chats without stating your own guess, and treat agreement as a starting point, not a verdict.
Can ChatGPT tell me what season I am?
It can give you a candidate season and a runner-up. The hard part is the same one people face: separating neighbors such as Soft Autumn and Soft Summer, where the difference is a small shift in warmth that a camera's white balance can erase. If repeated runs split between two neighboring seasons, compare those two directly instead of asking again.
What is the best ChatGPT prompt for color analysis?
One that fixes the procedure. Ask it to act as a 12-season analyst, describe and correct for the lighting first, read undertone at the jawline and neck, judge value from hair, brows, and eyes together, ignore clothing, and give a top season with a runner-up and the evidence for each. Tell it not to ask for your own guess. The full prompt is in the prompt section above.
Can AI find my undertone?
It can estimate it from a photo, and undertone is the call most exposed to lighting. Warm bulbs make cool skin look golden; shade makes warm skin look flat. A tool built for the job identifies the light first and asks calibration questions to check the photo against. You can also read undertone yourself with the skin undertone test, which needs no photo.
Are color analysis apps accurate?
They can be, on a clean photo. Every photo-based analysis shares the same input limit: indirect daylight, no filter, and no heavy makeup, or the read suffers. Look for an app that corrects for lighting, shows the evidence behind its verdict, and tells you when a second season is close instead of claiming certainty. Even a good one can be wrong: it can land on a neighboring season, or read cool coloring as warm when the light is warm.
Is it safe to upload a selfie to ChatGPT or a color analysis app?
It depends on the service, so check three things before uploading: what it keeps, for how long, and whether your photo is used to train models. Chatbots publish data controls in their settings. At Colorito, your photo is sent to our AI provider (Anthropic) for processing and stored securely with your result in your account. Stored photos are used to review and improve the analysis, are never sold or shown publicly, and are deleted on request. The privacy page has the details.
Is ChatGPT color analysis free?
It costs nothing beyond the plan you already have, as long as that plan accepts photo uploads. The price you pay is in effort and uncertainty: setting up a clean photo, writing a careful prompt, and repeating the run to see whether the answer holds.
Sources
- OpenAI: Images and vision guide: the published limitations of OpenAI's vision models: they can make mistakes, may generate incorrect descriptions, and may resize images before analysis
- OpenAI Cookbook: Reproducible outputs with the seed parameter: model outputs are non-deterministic by default and "may differ from request to request"
- Sharma et al. (2023), Towards Understanding Sycophancy in Language Models: five AI assistants consistently favored responses matching a user's stated views across four text-generation tasks
- Wikipedia: Color balance: white balance: how a camera adjusts an image's overall color so neutral objects look neutral
- Colorito: The guide to color analysis: the rules Colorito's analyst follows on every photo, under "How the Read Is Made"
- Colorito: What a color analysis costs: the US studio price index behind the in-person price band