AI-powered cosmetic skin analysis. No API keys. No monthly AI costs. Fully self-hosted.
A complete skin analysis system that sits behind your "Analyse My Skin" button on your Shopify store. Customers upload a selfie, receive a personalised cosmetic skin analysis with visual overlays, and are matched to their perfect CLEVE routine.
No Anthropic key. No OpenAI. No recurring AI costs. The AI model is trained on open datasets and runs on free hosting.
Customer clicks button → CLEVE modal opens → uploads selfie
→ EfficientNetV2 + MediaPipe + OpenCV analyses skin
→ 9 concerns detected with severity scores
→ Face map with coloured overlays generated
→ CLEVE products matched and recommended
→ Customer adds to Shopify cart
| Concern | Detection Method | Output |
|---|---|---|
| Acne (Whiteheads, Blackheads, Papules, Pustules) | EfficientNetV2-S | Severity 0-100, confidence % |
| Hyperpigmentation & Dark Spots | Neural network | Heatmap overlay |
| Redness & Sensitivity | Neural network + OpenCV | Severity, regions |
| Dark Circles | Neural network | Under-eye segmentation |
| Fine Lines & Wrinkles | Neural network | Severity by zone |
| Enlarged Pores | OpenCV high-frequency analysis | Density score |
| Skin Texture | OpenCV local std deviation | Roughness score |
| Acne Scars | Neural network | Severity, regions |
| Oiliness | OpenCV V-channel analysis | Shine estimate |
| Dataset | Source | Size | License | Covers |
|---|---|---|---|---|
| ACNE04 | Kaggle | 1,457 images | Academic | Acne severity grading |
| Acne Type Dataset | Kaggle | ~5,000 images | Apache 2.0 | 5 acne subtypes |
| Mendeley Skin Disease | Mendeley Data | 9,548 images | CC BY 4.0 | Acne, hyperpigmentation |
| Fitzpatrick 17k | GitHub | 16,577 images | CC BY 4.0 | Diverse skin tones |
| Roboflow Skin Detection | Roboflow | ~3,000 images | Free tier | Multi-label: pores, redness, dark circles |
| Component | Platform | Cost | Specs |
|---|---|---|---|
| AI Backend (FastAPI) | Hugging Face Spaces | Free | 2 CPU, 16GB RAM |
| Frontend (Next.js) | Vercel | Free | 100GB bandwidth/mo |
| CI/CD | GitHub Actions | Free | 2,000 min/mo |
| Total | £0/month |
See docs/DEPLOYMENT.md for the complete step-by-step guide.
TL;DR:
- Train model:
cd training && python train.py --train - Push
backend/to a Hugging Face Space - Push
frontend/to Vercel - Paste
shopify/cleve-skin-analysis-button.liquidinto your Shopify theme - Done ✓
This platform is a cosmetic analysis tool only.
- ✅ Uses compliant language: "visible signs consistent with..."
- ✅ Never claims to diagnose disease
- ✅ Always displays medical disclaimer
- ✅ No facial recognition or identity matching
- ✅ No biometric identification
- ✅ Explicit consent required before analysis
cleve/
├── training/ Model training pipeline
├── backend/ FastAPI backend (→ Hugging Face Spaces)
├── frontend/ Next.js frontend (→ Vercel)
├── shopify/ Shopify theme snippet
└── docs/ Deployment guide
"# Cleve-AI"