Built independently.
Skinzy is operated by Yash Ghansham Thakare, not a company with a marketing budget or an incentive to sell a specific brand. That's the whole reason it can recommend by ingredient instead of by sponsorship.
Why this exists
Most "AI skin analysis" tools online are a marketing front for one brand's product line — the diagnosis conveniently always ends at that brand's serum. Skinzy started from a different constraint: score the skin first, from what's actually visible in the photo, and only then figure out which ingredients address what was found — regardless of who sells them.
It's built for the Indian market specifically — Indian skin tones, Indian humidity and pollution conditions, and skincare actually available to buy in India, rather than a US or Korean product catalogue with a currency symbol swapped.
What it's trained on
Google's MediaPipe FaceMesh — 468 open-source facial landmark points, Apache 2.0 licensed.
A MobileNetV2 classifier fine-tuned on a labeled acne-severity dataset, run locally via ONNX — no image ever leaves the request to a third-party AI API.
Fine-tuned on Roboflow Universe's "FaceAttribute 2" dataset by HeheTeam, CC BY 4.0 licensed. Full attribution in the repository's ATTRIBUTIONS file.
What we deliberately don't claim
- This is not a medical diagnosis and does not replace a dermatologist. See Terms for the full scope.
- Pigmentation scoring is the weakest of the five axes — it measures light variance, which can be confounded by shadow and pore texture, especially on darker skin tones. We say this plainly rather than hide it behind a confident-looking number.
- Your selfie is never stored — see Privacy for exactly what is and isn't kept.