The problem AI face matching solves
A typical wedding generates between 1,500 and 4,000 edited photos. The couple gets them all. The 400 guests who attended each want to find the photos they appear in — but no one is going to manually sort 4,000 images by person.
Traditionally, the options were: the photographer exports by person (not realistic at scale), the host dumps everything in a shared folder and guests dig through it themselves (terrible experience), or the photographer sends individual guests their subset manually (possible for small events, impossible for large ones).
AI face matching eliminates this problem entirely. Each guest takes one selfie, and the system auto-delivers every photo that contains their face.
What is AI face matching?
AI face matching is a computer vision technique that identifies specific people across a large photo set. The system works in two steps:
First, it runs face detection across every image in the event — locating and cropping every face it finds. Then, it generates a vector embedding for each face: a mathematical representation of the facial geometry (distance between eyes, jaw width, nose bridge, etc.) compressed into a fixed-length array of numbers.
When a guest submits a selfie, the same embedding is generated for their face. The system then performs a vector similarity search — comparing the selfie embedding against all embeddings in the event — and returns photos where the similarity score is above a confidence threshold. The result is a filtered gallery containing only photos where that person appears.
How it works, step by step
Photographer uploads photos to Pixbox
The full shoot — 2,000 to 4,000 images — gets uploaded to the event album. Upload can happen during the reception, the day after, or whenever editing is complete.
AI processes faces and generates embeddings
Pixbox runs computer vision across every photo, detecting faces and generating a mathematical vector (embedding) that represents each face's unique geometry. This happens automatically in the background — no manual tagging required.
Guest opens the album link on their phone
The host shares one link. No app install required. The guest opens it in their browser — works on iOS and Android without downloading anything.
Guest takes or uploads one selfie
The guest takes a selfie with their front camera, or uploads a recent photo. The selfie is processed in-session: a face embedding is generated the same way as the event photos.
Pixbox matches against all 3,000 photos
Pixbox runs a vector similarity search: the selfie embedding is compared against every face embedding in the event. Photos where the similarity score exceeds a confidence threshold are returned as matches.
Guest sees only photos with their face in under 30 seconds
The guest's personal feed loads — filtered to only the photos they appear in. They can browse, download, or share directly from the gallery. No waiting for the photographer to manually sort.
What AI face matching can and cannot do
Can do
- Match faces despite different lighting conditions (harsh outdoor flash vs. soft indoor)
- Match when a guest is wearing glasses in some photos but not others
- Match from slight angle variations (profile vs. straight-on)
- Handle events with 400+ unique guests across 3,000+ photos
- Return matches in under 30 seconds for typical event sizes
Cannot do
- Match if a face is obscured more than 50% (face buried in bouquet, facing away)
- Reliably match toddlers or very young children (facial geometry changes too quickly)
- Identify people who are not actually in any photos
- Match with high confidence when the selfie is very low resolution or taken in near-darkness
The privacy angle
A common concern: are selfies stored as permanent biometric data?
In Pixbox, selfies are processed in-session. The image is used to generate an embedding and then discarded — only the embedding (a mathematical vector, not a photo) is retained for the matching operation. Guests do not create accounts. Their selfie is not linked to a persistent identity profile.
The face embeddings generated from event photos are stored per-event and scoped to that album — they are not cross-referenced against other events or external databases. This design was chosen specifically to maintain GDPR-safe operation and avoid the legal and reputational risks of building a centralized biometric registry.
What this means for the host
The host sets up face matching by enabling it when they activate the event. After that, it is automatic. Guests receive the album link (the same one that handles RSVPs and the event schedule on Guestcard), tap to find their photos, and within 30 seconds they have a personal gallery of every photo they appear in.
The practical effect: guests stop texting the couple asking for photos. The couple stops fielding those messages. The photographer does not need to manually sort anything. The reception experience changes — guests are browsing their own photos on their phones while the event is still happening.
AI face matching ships with every Standard and Premium event
No add-on pricing, no separate AI tier. Included in Standard ($59/event) and Premium ($149/event) activations.
See the feature in detail