Region-by-region analysis
T-zone, cheeks and chin are read separately, each with its own assessment — not one score for the whole face.
SKINLENS · AI SKIN REFERENCE ASSISTANT
One selfie, four skin dimensions
A vision LLM analyzes your T-zone, cheeks and chin region by region, mapping you to one of 16 skin types with a four-dimension spectrum report and care suggestions.
Results are AI-generated, for skincare reference only — not medical advice.
Features
More than a single vague score
From per-region reads to a four-dimension profile, every step comes with evidence and a confidence level.
16-type spectrum report
Four bipolar spectrums — oil, sensitivity, acne, pigmentation — each with a confidence level; low-confidence results are clearly marked as tentative.
Care suggestions
Daily care tips derived from a 16-type skin handbook, always phrased as reference and suggestions — no efficacy claims.
Privacy first
Photos are deleted immediately after analysis and never stored long-term; your history lives only on your device.
How it works
Three steps to your report
- 1
Take a front-facing photo
Natural light, no makeup, full face visible. The Android app offers an in-page live viewfinder.
- 2
AI analysis
Input is quality-checked first: non-face photos, screen re-shots and distant shots are rejected with retake guidance — no misleading reports.
- 3
Read your report
Spectrum report + per-region assessment + care suggestions. Save locally and revisit anytime from the “Me” tab.
Guides
Understand your report, starting here
Five guides covering photo prep, how the analysis works, confidence levels, privacy and skincare choices. Currently available in Chinese.
How to take the selfie
Bare skin, natural light, distance and sharpness — plus a table of common failure cases.
The four dimensions & 16 types
How oil, sensitivity, acne and pigmentation combine into 16 skin types, with a Baumann comparison.
How AI photo skin analysis works
Quality gate, region-by-region reads, four-dimension calls — what it can and cannot see.
What confidence levels mean
Why low confidence is marked tentative, and the delete-after-analysis privacy boundary.
Choosing skincare from your result
Cleansing, sunscreen and moisturizer by dimension — ingredient and category level only, zero brands.
Download
Two ways to start
The web version needs no install; the Android app adds a live viewfinder.
Web (H5)
Open in your mobile browser — no install, full functionality.
Try nowAndroid APK · Flutter
In-page live viewfinder capture. Signed with a test certificate — allow “unknown sources” when installing.
DownloadAndroid APK · uniapp
Native build sharing the H5 codebase, with width-capped tablet layout. Signed with a test certificate — allow “unknown sources” when installing.
DownloadTech
One backend, two stacks, four targets
Also a cross-platform engineering sample: uniapp and Flutter frontends replicate the same API and visual language, with single sources of truth for both the data contract and design tokens.
uniapp · Vue3 H5 / WeChat MP / App (APK)
Flutter Android APK (live viewfinder)
Hono unified API · dual-cloud deploy: Alibaba Cloud FC (mainland direct) + Cloudflare Workers (mini program / fallback)
D1 / R2 · reports & transient images (Workers path, deleted after use)
Qwen VL · multimodal analysis (same-region from FC)
Abuse guard · per-IP rate limit + CORS allowlist
Single-source data contract
The report structure is defined once in JSON Schema; scripts generate TypeScript and Dart types, and LLM output is validated against the same schema.
Design-token SSOT
One DTCG design-tokens.json generates SCSS variables and Dart constants — zero visual drift between stacks.
Dual-cloud deploy, one codebase
One Hono app ships to two clouds via a platform adapter: mainland traffic hits Hangzhou FC directly (analysis ~35s → ~5s), the mini program stays on Cloudflare Workers; the API ships per-IP rate limiting and a CORS allowlist.
AI engineering
Input quality gating runs inside the same vision call: non-face / re-shot / too-far photos get a 422 with retake guidance and are never persisted.
Privacy by architecture
Images are deleted after use with an R2 lifecycle rule as backstop; history is written only to local device storage.
Per-stack strengths
The Flutter build adds an in-page live viewfinder (camera plugin) with full fallback to the system camera; the uniapp build wins on one-codebase-many-targets.
Compliance & privacy
Boundaries, stated upfront
Disclaimer
Results are AI-generated and for skincare reference only. They do not constitute medical advice — please see a doctor for serious skin conditions.
Privacy promise
Uploaded photos are used only for the current analysis and deleted as soon as it completes; your history is stored only on your device — no personal profile is kept server-side.