Fair skin, ash-brown hair, grey-blue eyes
A warm lamp turned a Soft Summer into a Soft Autumn: a cool palette became a warm one.
Our test ·
ChatGPT color analysis:we tested it 27 times
Short answer: an OpenAI GPT model will name your color season from a selfie, and in daylight it gives the same season run after run. Change only the light, and it changed the season for two of our three test faces: a cool Soft Summer became a warm Soft Autumn under a lamp.
The results
Same face, different light, different season
Each photo was sent three times, each time in a fresh session. The label under a photo is the season it got; the dots show the confidence the model gave itself on each run.
Medium olive skin, dark brown hair and eyes
Fluorescent light turned a Deep Autumn into a Deep Winter, every time.
Deep brown skin, black coily hair, very dark eyes
Stable in daylight; under fluorescent light one run in three said Deep Winter instead.
Filled dot = medium confidence, empty dot = low. No run reported high confidence. Pink = a different season from the daylight photo. Undertones listed per run.
What we found
Is ChatGPT color analysis accurate?
26 of 27
runs gave the same season as the other runs on that photo. Asked twice about one photo, it rarely contradicts itself.
2 of 3
people got a different season once the light changed. A cool palette turned warm under a lamp, and a warm one turned cool under fluorescent light.
4 of 9
photos got a different undertone label (warm, cool or neutral) between runs, even when the season stayed the same.
16 of 18
runs under lamp or fluorescent light rated themselves low confidence, yet every one still named a season.
The model was honest about its limits in words. Under the warm lamp it wrote that “the strong golden lighting makes your natural undertone difficult to judge”, and then answered Soft Autumn anyway. That is the real risk of a chatbot verdict: it sounds reasoned, and a single answer in bad light can send you shopping for the opposite palette.
Daylight helped most. All nine daylight runs were rated medium confidence and agreed on the season. What none of the runs can tell you is whether that daylight season is right, which is why we measured consistency, not accuracy: these are generated portraits with no draped “true” season to compare against.
How we tested
Our method
- Photos: three AI-generated people, no makeup, plain grey t-shirt, off-white wall. Each was photographed in daylight, then edited so that only the lighting changed: a warm tungsten lamp and cool fluorescent light. No real person’s photo was used.
- Model: OpenAI’s gpt-6.1-sol, through OpenAI’s Codex command-line tool, with tools disabled and a fresh session for every run. This is an OpenAI GPT model answering a photo question, not the ChatGPT app itself, so your results in ChatGPT may differ.
- Runs: 9 photos × 3 runs = 27 answers on . Every answer is shown above, including the low-confidence ones.
The prompt we used
This is a selfie of me. Do a personal color analysis from the photo and tell me my color season in the 12-season system. Reply in exactly this format: SEASON: <one of the 12 seasons> UNDERTONE: <warm, cool or neutral> CONFIDENCE: <low, medium or high> REASON: <one sentence>
If you still want to try it
A better ChatGPT prompt for color analysis
Our results point to four fixes: control the light, skip the filters, give it more than one photo, and make it show its reasoning before it commits to a label.
- 1Face a window in daylightNo lamps, no ring light, no flash. Light was the only thing we changed, and it flipped the season twice.
- 2Turn off filters and beauty modesPortrait and beauty modes warm or smooth the skin, which is exactly what the model reads.
- 3Send three photos, not oneDifferent photos expose the white-balance guesswork. If the answers disagree, the photos are the problem.
- 4Ask for two candidatesA shortlist you can test with real colors under your face beats one confident label.
Copy this prompt
I'm attaching three unfiltered selfies taken facing a window in daylight, no makeup, hair visible. Before naming a season, describe separately: 1. My skin undertone (warm, cool, neutral or olive) and what in the photos shows it 2. My value (light, medium or deep) across skin, hair and eyes 3. My chroma (clear and bright, or soft and muted) 4. The contrast between my hair, eyes and skin Then give your top two 12-season candidates, what would decide between them, and which colors I could hold under my face to test it. Tell me if the lighting makes any of this unreliable.
We wrote this prompt from what the test showed; it was not part of the 27 runs.
Your options
ChatGPT vs a quiz vs a color analysis
| ChatGPT | Free quiz | Palette Hunt analysis | |
|---|---|---|---|
| Input | Any photo you upload | Answers about your hair, eyes and skin | Guided photos |
| Method | Changes with your prompt and chat | Same scoring for everyone | Same method on every analysis |
| Result | A season name and a short explanation | One of the 4 main seasons | 12-season result, palette, colors to avoid, makeup and hair guidance |
| See colors on you | No | No | Recommended colors shown on your photo |
| Cost | Free with limits | Free, no signup | One-time payment |
Palette Hunt publishes this page. A photo analysis still depends on the photos you give it, so take them in daylight too.
Questions, answered
ChatGPT color analysis FAQ
Can ChatGPT do color analysis from a photo?
Yes, it will look at a selfie and name a season, and in good daylight its answer is usually consistent. But it reads the colors in the photo, not your skin: in our test, changing only the lighting changed the season for two of three people, and the model still answered with a season even when it said the lighting made the undertone hard to judge.
Is ChatGPT color analysis accurate?
Treat it as a first guess. Asked about the same daylight photo three times, the season matched in all nine daylight runs, but the undertone label changed between runs for two of the three people. Under a warm lamp or fluorescent light, the verdict moved to a different season family for two of three people.
What is the best ChatGPT prompt for color analysis?
Upload several unfiltered daylight selfies and ask it to describe undertone, value, chroma and contrast before naming a season, then to give its top two seasons and what would decide between them. That turns one confident label into a shortlist you can check with real fabrics or colored paper under your face.
Why does ChatGPT give me a different season each time?
Each chat starts fresh, and the model is reading the color of the light as much as your skin. In our test a warm lamp turned a Summer into an Autumn, and fluorescent light turned an Autumn into a Winter. Different photos, filters and phone white balance make it worse.
Is ChatGPT color analysis free?
Asking costs nothing beyond your ChatGPT plan and its usage limits. What you get back is a text answer: a season name, a few color suggestions and an explanation, not a full palette or a preview of colors on your face.
ChatGPT or a color analysis app: which should I use?
ChatGPT is good for learning the vocabulary and getting a starting guess. A dedicated analysis gives you a fixed method, a full palette and colors to avoid, and Palette Hunt also shows recommended colors on your own photo. A free quiz is a third option if you want a direction without uploading anything.
Keep going
Free color analysis quizFind your season direction without uploading a photo.Skin undertone testCheck warm, cool or neutral with the vein, jewelry and white-paper tests.The 16 color seasonsHow the 12 and 16 season systems split each season.Best color analysis apps8 tools compared by price, input and result.Soft Summer color paletteThe season our fair-skinned test portrait got in daylight.Deep Autumn color paletteThe daylight verdict for our other two test portraits.
Get a season you can rely on
Upload your daylight photos for a full 12-season analysis, palette and colors shown on your own face.
Start my analysis