MoleCheck live demo
The same model that powers the MoleCheck application, running in your browser. Your photo is analyzed on your device — it is never uploaded, and this site has no server to receive it.
Before you begin — this is not a medical test.
- This is an educational demonstration of image classification. It cannot diagnose skin cancer or any other condition.
- A "lower-risk" result is not reassurance. The model still misses roughly 1 in 10 malignant lesions even on its test set, and a real lesion can look unlike anything in its training data.
- If a mole is new, changing, bleeding, or concerning in any way, consult a dermatologist — regardless of what this page reports.
Drop a close-up photo of a single mole here,
or click to choose one
Whatever the score: this tool cannot perceive what a dermatologist can. If you are concerned about this mole, have it examined in person.
What happens to my photo?
Nothing leaves your browser. The model (a 21 MB neural network) is downloaded to your device once, and all inference runs locally via WebAssembly. You may disconnect from the internet after the page loads and it will continue to function.
Tips for a meaningful result
- One mole per photo, centered and in focus, filling most of the frame.
- Even lighting, with no shadows across the lesion.
- This is the v2 model, trained on real smartphone photos of lesions — so an ordinary phone photo is now in-distribution. It is still only an educational signal, not a diagnosis.