尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
正在載入完整 README。 以下為已儲存的摘錄。
Upload a photo and find out which of the 12 colour seasons you are. You get your palette, makeup shades, jewellery and hair colours, and you can check clothes against your palette before you buy them.
1. You upload one photo. It is checked in your browser first (lighting, focus, one face). If it fails, nothing is uploaded.
2. Your skin, hair and eye colours are measured on your device in CIE Lab.
3. The measurements are scored against the 12 seasons. A vision model looks at the photo and picks the final season from the top three.
4. You get four tabs: Overview, Beauty, Style and Shop.檢視完整 README ↗Upload a photo and find out which of the 12 colour seasons you are. You get your palette, makeup shades, jewellery and hair colours, and you can check clothes against your palette before you buy them.
Live: https://aura-azure-six.vercel.app
The palettes, makeup shades, gemstones and hair colours are fixed lists per season. The model picks the season and names the looks, but it never invents a colour.
The 12 seasons
What you get
Upload
Analysing
Results
Makeup looks
Style
Before you buy
On a phone
@jsquash WASM image decodersYou need Node 22 or newer and an OpenRouter API key.
npm install
npm install --prefix client
cp .env.example .env # then add your OPENROUTER_API_KEY
npm run dev # site on localhost:5173, API on localhost:3001
An analysis costs about $0.01 in model credit. The sample faces are free.
npm test # unit tests
npm run e2e # full flow in a real browser (uses model credit)
npm run e2e:home # landing page checks, free
npm run shots # screenshots of every screen into dev/shots
npm run diagnose -- me.jpg # show the measurements and ranking for a photo
client/ the React app
api/ Vercel Functions, one file per route
server/ the route handlers, prompts and season data
measure/ the colour measurement code (runs in the browser)
eval/ accuracy testing
tests/ unit tests
scripts/ maintenance and dev tools
Photos are measured in the browser. The image that is sent for analysis is kept in memory only, never saved and never logged. Results are stored in your own browser, so a results link only opens on the device that made it.
To keep costs under control each visitor gets 5 analyses, 10 shop checks and 15 chat messages a day.
This is a personal project and the accuracy testing is still in progress. The sample faces are AI generated. CLAUDE.md has the architecture notes and a log of what was measured and changed.
MIT licence. Made by Haneen.