How it works
Pick a folder
Choose any directory on your machine to scan.
Find duplicates
Images are embedded via CNN and matched by cosine similarity.
Review & clean up
See grouped results, keep the best, move the rest to Trash.
Under the hood
Each photo is run through a small CNN (MobileNetV3) locally on your machine to produce a feature embedding — a vector that captures the visual content of an image.
These vectors are L2-normalized and indexed with FAISS (Facebook AI Similarity Search). Cosine similarity between any two vectors determines how alike they are. Pairs above a threshold are grouped as duplicates using Union-Find.
Features
Sensitivity control
Tune matching strictness — from near-exact copies to visually similar photos.
Quality ranking
Surfaces the best version per group by resolution, sharpness, and metadata.
Autocrop
Strips borders and whitespace before comparing so padded images still match.
Safe by default
Deletions go to Trash. Nothing is permanently removed.
Fully local
Runs entirely on your machine. No cloud, no uploads, no internet required.
HEIC support
iPhone photos just work — HEIC, JPEG, PNG, WebP, and more.