Personal color analysis, done properly.
Five calibrated photos. Four dimensions. One honest verdict.
Most AI color tools return a season in ten seconds from one bad selfie. True Hue doesn't. It runs the protocol a certified analyst runs when you pay for a real consultation: calibrated photo intake, four-dimension analysis with receipts, a mandatory pause before any verdict, and adjacent-season draping. If the inputs aren't good enough, you get a reshoot request — which is better than a confident wrong answer.
/plugin marketplace add https://github.com/bankielewicz/TrueHue /plugin install true-hue@truehue-marketplace
The difference between analysis and a party trick.
You take one selfie in your bathroom at 10pm. The app thinks for two seconds. "You're an Autumn!" Cool. That's a fortune cookie with a color filter on it.
Real seasonal color analysis is a craft. Trained analysts spend years getting good at it. They read skin in calibrated light from three different angles. They drape fabric at your shoulders to see what actually changes in the mirror. They'll tell you honestly when you're sitting on a border between two seasons, because forcing a verdict on an ambiguous case makes your palette worse, not better.
True Hue brings that protocol to Claude Code. It refuses to work from one bad photo. It asks the questions a human analyst would ask: foundation history, compliment patterns, what colors consistently made you feel invisible. It shows its reasoning, photo by photo, answer by answer.
Halfway through, it pauses. Before a season gets named, it presents its four-dimension reading back to you with two targeted questions designed to surface disagreement — not confirmation. Push back with evidence and the reading shifts. Push back with just preference and Claude holds the line and explains why. This phase exists specifically to create an exit ramp for wrong conclusions.
Deep skin and olive complexions aren't handled as afterthoughts. Color-blind users can opt into an accessible deliverable. Privacy guidance comes before the photo request, not buried three scrolls down. None of this is ornament. All of it is the difference between a palette you can actually shop and a season you were handed like a horoscope.
A full deliverable, in about thirty minutes.
Not a season name and a shopping list. A working palette tailored to you, with the reasoning, the context, and the caveats.
- Your season, with honest confidence level
- Four-dimension analysis, photos and answers cited
- Why not each adjacent season, specifically
- ~36 hex codes organized by wardrobe role
- Within-palette pairing rules
- Fabric sheen and weight guidance
- Pattern and print recommendations
- 8–12 colors to avoid, each with its reason
- Your specific neutrals: "your black," "your navy," "your denim"
- Metals: primary, secondary, avoid near face
- Grooming sections — only the categories you picked
- Starter capsule at two budget tiers, with six-month phasing
Four families. Twelve temperaments.
Spring is warm and bright. Summer is cool and muted. Autumn is warm and muted. Winter is cool and bright. Each family splits into three sub-seasons — one leaning toward each neighbor on the flow chart.
One swatch per season — a LAB-validated anchor from the Hue Atlas reference set. Your actual deliverable contains the full ~36-color working palette for your season, calibrated to your specific within-season position. A Soft Autumn near the Warm Autumn border gets richer mids. A Soft Autumn near Soft Summer gets cooler neutrals. That kind of specificity is the difference between a palette you can shop and a list of adjectives. Full reference in the twelve_seasons doc.
Seven phases, about thirty minutes.
Plus photo-capture time, which runs ten to fifteen minutes on its own. The first phase is just routing questions. The last three are where the actual palette happens.
-
Consultation prep
Three routing questions up front — have you been typed before, what are you trying to get out of this, how much time do you have. Point is to not drag you through a half-hour intake when a five-minute single-item check is what you actually came for.
-
Intake protocol
Five photos in natural indirect daylight — a north-facing window works, an overcast day outside works. Plain white printer paper held at your cheek as a white-balance reference. Questionnaire covers hair (natural and current), sun response, jewelry harmony, compliment patterns, foundation history for anyone who wears it, lifestyle and climate, plus the grooming categories that apply to you.
-
Quality gate
Claude inspects the photos against the protocol. Flash artifacts, visible makeup, a reference paper reading yellow because a bathroom bulb was in frame, any missing shot — sent back for a reshoot. Usually just one photo needs redoing. If a clean reshoot genuinely isn't possible, there's a degraded-mode path with explicit confidence reduction.
-
Four-dimension analysis
Undertone, value, chroma, contrast — each worked through separately, with evidence cited from specific photos and specific answers. Deep skin gets the vein test skipped in favor of fabric draping and foundation behavior. Olive complexions get the green cast acknowledged, not forced into warm or cool.
-
Mandatory check-in
Before a season is named, the four readings come back to you with two targeted questions pulled from a vetted library. Bring evidence and the reading shifts. Bring just preference — "I want to be a Spring" — and Claude holds the line and explains why. This phase exists because misreads are cheap to fix here and expensive to fix downstream.
-
Draping & deliverable
Season named, with explicit "why not" reasoning for both neighbors. Then the full deliverable: ~36 hex codes by role, colors to avoid with reasons, metals, your neutrals specifically, pattern and fabric guidance, a two-tier starter capsule with a six-month phasing order, and the grooming sections you actually selected.
-
Walkthrough & follow-up
Optional PDF swatch cards in three formats if you want something pocketable. Walkthrough of any section in more depth. And if colors stop working after a few weeks of wear, there's a post-delivery validation path for adjustments within the season — or, rarely, a full reconsideration.
Things people ask, mostly.
Is this actually different from free color analysis quiz apps?
Two concrete differences. The quiz takes one selfie or asks you to pick eye and hair colors from a grid — which isn't enough signal to land a season reliably, and the math on it is shallow. True Hue requires five calibrated photos plus a fifteen-item questionnaire and works through four dimensions separately, each with cited evidence. Second: the quiz always hands back one answer in ten seconds because "between seasons" or "I can't tell" doesn't feel like a product. True Hue rejects bad inputs, acknowledges genuine ambiguity, and will tell you plainly when an in-person analyst would produce a better result than it can.
How long does it really take?
Plan on 20–30 minutes for a first full Mode 1 analysis. Roughly 10–15 minutes for the photos and questionnaire, plus 10–15 minutes of back-and-forth. Single-item checks are 2–5 minutes. Closet audits scale with how many items. Palette refinement — for people already typed who want to push borders — is 5–15 minutes.
What about deep skin? The standard Sci\ART tests don't work for me.
They mostly don't, and the plugin knows. The vein test gets skipped entirely — veins aren't reliably visible on deeply pigmented skin, and pretending otherwise wastes everyone's time. The sun-response test gets de-weighted for similar reasons. Undertone is determined instead by the white-vs-cream fabric drape, foundation oxidation behavior, compliment patterns, and jewelry harmony — all of which work across skin depths. Throughout the deliverable, deep skin is first-class: the foundation brand recommendations cover ranges that actually cover deep skin, the blush guidance explicitly notes higher saturation is needed to read visibly, and the protocol is built in rather than patched on.
Is my photo data safe?
Photos live in the chat session. When you delete the conversation, they're gone from future sessions. Anthropic does not use chat content, including images, for training by default — current policy is published on Anthropic's site. You can crop photos tight to show only the face area and skip the identifying background, logos, and name tags without affecting analysis quality. Anything you screenshot and share elsewhere is outside Claude's privacy envelope. (These statements apply to Anthropic's direct products; third-party apps built on the Claude API may have their own policies.)
I'm colorblind. Does this work for me?
Yes. The intake asks about color vision. If you indicate any form of deficiency, the deliverable includes a color-blind accessible format — standardized color names, value tiers, saturation levels, hex codes for reference — plus flagged palette pairs that may look similar to you under your specific type of deficiency, and shopping strategies that don't depend on your screen rendering colors correctly.
What if Claude gets the season wrong?
The mandatory mid-analysis check-in is designed to catch this before any season gets named. If something in the four-dimension reading feels off, push back with evidence and the reading shifts. If you only discover issues later — you've worn some of the palette and specific colors are failing — there's a post-delivery validation workflow. Most cases resolve with minor hex substitutions within the same season. A full reanalysis with fresh photos is rarer but sometimes the right move.
Does this work on the Claude web or mobile apps?
The plugin installs into Claude Code specifically. The underlying skill can run in any Claude environment with file read and write, but the install commands on this page are Claude Code commands. If you're on claude.ai, you can ask Claude to follow the intake and analysis approach described in the repo — it won't auto-load the skill but will work from the descriptions.