Guide
How face recognition picture sharing works
·5 min read
Face-based picture sharing can feel like magic: take a selfie, and every picture you’re in appears. Here’s what’s actually happening under the hood — in plain English.
Step one: detecting faces
When an event gallery is uploaded, the system scans every picture and detects the faces in it — including faces deep in the background or in a crowd. Each detected face is turned into a compact numerical representation (an “embedding”) that captures what makes it distinct.
This is what lets the system later tell whether two faces, in two different pictures, belong to the same person.
Step two: grouping by person
Faces with similar embeddings are grouped together, so all the pictures of one person cluster into a single set — without anyone naming or tagging them. A large event becomes an organized index of people rather than an undifferentiated pile of images.
Step three: matching your selfie
When a guest takes a selfie, it’s turned into an embedding the same way and compared against the gallery. The closest matches are the pictures that guest appears in — returned instantly as a personal gallery to view and download.
That’s the whole trick: detect faces, group them, and match a selfie against the group. With Picsort, all the guest sees is “take a selfie, get your pictures.”
What about privacy?
Good face-based sharing keeps each event private. Picsort albums are reachable only by the link the host shares — they aren’t publicly listed or indexed by search engines — so the people who find themselves are the people who were invited.
Frequently asked questions
Do I need to tag anyone for face recognition to work?
No. The system groups faces automatically, so no manual tagging is required. A guest simply takes a selfie to find the pictures they’re in.
Are the pictures and faces kept private?
With Picsort, each album is private and only reachable by the link the host shares — it isn’t publicly listed or indexed, so only invited guests can find their pictures.