All posts
GuidesOct 11, 2026 · 11 min read

How AI Face Recognition Photo Sharing Works at Events

AI face recognition photo sharing turns a selfie into numbers and matches it to event photos. Learn the steps, the common misses and the privacy laws.

A guest holds a phone up for a selfie at an outdoor party under string lights.
AI-generated illustration

AI face recognition photo sharing lets a guest find their own photos from an event with one selfie. The system turns the selfie into a list of numbers, compares that list with the faces in the event photos, and shows the photos that score above a set threshold.

This guide explains each step, why some photos are missed, what the privacy laws say in the EU, India, Illinois, Texas and Washington, and which questions to ask any vendor. ShotKite is one option among several. A later section shows how it handles selfies and deletion, so you can compare it with other tools.

How AI face recognition photo sharing works, step by step

Face recognition photo sharing follows a standard face matching pipeline, and a failure at any step can cause a missed photo.

  1. Detect the face. The software finds faces in each image and crops them. A selfie usually shows one large face. A wide event photo can hold many small faces, and each one goes through the same steps.
  2. Build a template. The cropped face becomes a template, a fixed-length list of numbers that describes it. A template is not a photo, but it may still count as biometric data under some of the laws covered below.
  3. Compare templates. The guest's template is compared with the templates of faces in the event photos. In the FaceNet approach, distance between templates tracks similarity, so a smaller distance means the faces look more alike.
  4. Apply a match threshold. Scores above a set threshold count as matches, and scores below it do not. A higher threshold misses more true matches, and a lower one lets in more false matches. NIST describes both error rates as functions of the same threshold, so tuning trades one against the other.
  5. Show the matches. Every match above the threshold is saved. The guest's gallery shows new photos while the page is open, through a live connection or a refresh every 10 seconds.

What the guest's selfie becomes

For the guest, the process starts and ends with one selfie. The selfie becomes a template, and that template is used to search for that guest in the event's photos. Two questions matter most: how long the template is kept, and which events it can reach.

A template limited to one event is easier to explain to guests and easier to delete. A template reused across many events raises harder questions, so ask about this before you choose a tool.

What guests see when they look for their photos

Guests who search for ways to find my photos from an event want one thing: the pictures where they appear. They take one selfie in their phone's browser, with no app download and no account. Once the match runs, the gallery shows those photos and keeps updating while the guest has the page open.

Phone screen of a guest gallery reading Found 8 photos of you, with Live updates on.
A matched-photos gallery on a guest's phone, which updates live while the page is open and refreshes every 10 seconds.Screenshot

Why some photos are missed, and how to fix it

A missed match is a false non-match: the system fails to link a guest to a photo that shows them. A 2020 survey of face recognition under occlusion says systems still perform far from satisfactory under large pose variation, varying illumination, low resolution, different expressions and occlusion.

  • Lighting and contrast. A May 2025 preprint on synthetic faces found that false negatives rise as contrast or brightness degrade. Take selfies in even light on the guest's face, not with a bright window or the sunset behind them.
  • Angle. The same preprint found that false negatives rise as the head's pose shifts. Ask guests to face the lens straight on for the selfie.
  • Motion blur and resolution. The preprint found that blur and low resolution had especially strong effects. Ask guests to hold still for the selfie, and photographers can use a faster shutter speed for moving subjects.
  • Sunglasses, hats and masks. Anything that covers part of the face counts as occlusion. NIST's 2020 tests found that every algorithm tested had more missed matches on masked faces. For the most accurate algorithms, the miss rate rose from about 0.3% to about 5% under the heaviest mask coverage. Those were one-to-one checks, not event searches, so read the numbers as a direction, not a forecast. Ask guests to remove sunglasses and masks for the selfie.
  • Crowds and small faces. In a wide shot, faces can be small, and the preprint found that low resolution raised misses. Hosts running conference photo sharing should plan closer shots for the networking break.
  • Children. Accuracy for children's faces is not well documented in the sources reviewed. Ask any vendor for child-face test results before you promise a match to a young guest.
Attendees chat with coffee cups in a bright hall during a conference networking break.
Networking breaks are dense and fast-moving, which is where wide shots and small faces make matching harder.AI-generated illustration

What hosts and photographers can do before guests arrive

Several of the misses above come from conditions you can control. A short test before guests arrive catches the obvious problems.

A wedding photographer in a dark jacket raises a camera while guests dance under warm lights.
Good light at the venue helps both the photographer's shots and the guests' selfies.AI-generated illustration
  • Test a few selfies at the venue, at the spot where guests will stand, in the light they will have.
  • Choose a selfie spot with even light, and avoid a backdrop that is brighter than the guest's face.
  • Print a permanent QR code at eye level near that spot, so guests can scan it and take the selfie there.
  • Tell guests in one sentence what the selfie is used for, and point them to a privacy notice before they start.
  • Decide now who answers deletion requests and how fast you will act. Under India's data protection law, according to the PIB explainer, a response is due within ninety days once those duties apply.

Privacy and face data: the laws to know

Not legal advice

This section is general information about the laws, not legal advice. Check with a lawyer before you rely on it for your business or your event.

LawWhereWhat it requires
GDPR Article 9European UnionProhibits processing biometric data to uniquely identify a person, unless an exception applies. Explicit consent is one exception (Art. 9(2)(a)). Member states may add further conditions (Art. 9(4)).
Digital Personal Data Protection Act, 2023 and DPDP Rules, 2025IndiaNotice, and consent that is free, specific, informed, unconditional and unambiguous (s.5, s.6). Withdrawal as easy as giving consent (s.6(4)). Verifiable parental or guardian consent for under-18s (s.9). Erasure after withdrawal (s.8(7)). Phased in; see below.
Tex. Bus. & Com. Code 503.001Texas (CUBI)Inform the person and get consent before capturing a face geometry or other biometric identifier for a commercial purpose. Destroy it within a reasonable time and no later than one year after the purpose expires, subject to exceptions. Civil penalty up to $25,000 per violation, recoverable by the Attorney General.
740 ILCS 14, BIPAIllinoisPublish a written retention and destruction policy, made available to the public (s.15(a)). Inform the person in writing before collecting the identifier (s.15(b)). People harmed by a violation can sue (s.20). Photographs are excluded from the definition (s.10).
RCW 19.375WashingtonBefore enrolling a biometric identifier in a database for a commercial purpose, give notice and get consent, or provide a way to prevent later commercial use (19.375.020(1)). Keep retention to what is reasonably necessary (19.375.020(4)(b)). The Attorney General enforces it (19.375.030(2)).

Under GDPR Article 4(14), biometric data is personal data that results from specific technical processing and allows unique identification, with facial images given as an example. Photos are treated differently: Recital 51 says processing photographs should not automatically count as special category data, unless a specific technical means allows unique identification.

Illinois (740 ILCS 14/10) and Washington (RCW 19.375.010) both exclude photographs from their definitions. Whether a face template built from a photo falls outside that exclusion is a question for counsel.

India: the DPDP Act and the 2025 Rules

The Digital Personal Data Protection Act, 2023 received presidential assent on 11 August 2023. The DPDP Rules, 2025 were notified on 14 November 2025, and the government's PIB explainer says they give the Act full effect, with an eighteen-month period for phased compliance. According to dpdpa.com's enforcement timeline, the main duties, including notice, consent, children's data and rights requests, start around May 2027. Check the official notifications before you rely on that date.

Once the consent rules apply, each consent request must come with a notice, and consent must be free, specific, informed, unconditional and unambiguous (the Act, s.5(1) and s.6(1)). A data fiduciary must be able to prove that notice was given and consent obtained (s.6(10)). According to the PIB explainer, penalties can reach Rs 250 crore for failing reasonable security safeguards, once these duties apply.

Children need extra care. The Act defines a child as anyone under 18 (s.2(f)) and requires verifiable parental or guardian consent before processing a child's data (s.9(1)).

For camp photo sharing, school events and youth sports, plan that consent step in advance. The Act also has a personal or domestic purpose exemption (s.3(c)(i)), and you should ask a lawyer whether it covers guest uploads to a shared gallery.

How ShotKite handles guest selfies

  • Each face template belongs to one event. A guest is matched only against the photos from the event they joined.
  • The selfie becomes a 512-number template. ShotKite keeps the template for matching, not the selfie photo.
  • The selfie screen says the selfie is used only to find that guest in that event's photos, and links to a guest privacy page.
  • Guests need no app and no account. The selfie is taken live in the phone browser, not picked from the camera roll.
  • A guest who joins late still sees earlier photos they appear in, and the gallery keeps updating while the page is open.
  • A photographer can delete a guest, which also removes that guest's matches, or a single photo. Guests can ask for deletion by email.

The full terms are on the privacy page.

Questions to ask any vendor

Before you choose an AI photo sharing tool, ask the vendor these questions in writing. Any face recognition photo app or photo sharing app with face recognition should answer them. For a wider comparison, read the best AI photo sharing apps for event photographers.

  1. Are templates scoped to one event, or reused across events and for other purposes, including model training?
  2. Do you keep the selfie image, the template, or both? For how long?
  3. Can a guest or the photographer delete the template and its matches, and how quickly does that happen?
  4. What notice and consent does a guest see, and can you prove that consent was given?
  5. How do you handle guests under 18, and how do you obtain verifiable parental consent?
  6. What test results do you have for masks, sunglasses, low light, motion blur and children, and which threshold do you use?
  7. Where is the data stored, and which laws do your terms rely on?

Frequently asked questions

Does AI face recognition photo sharing find every photo of a guest?
No. Misses happen with poor lighting, sideways angles, masks, sunglasses, motion blur and small faces in wide shots. The fixes above help, and a vendor should show test results for these conditions before you rely on it.
Is my selfie kept after matching?
It depends on the vendor, so ask whether the selfie image, the template or both are kept, and for how long. Under India's DPDP Act, once its duties apply, withdrawing consent must be as easy as giving it, and personal data must be erased after withdrawal unless another law requires keeping it.
Is face recognition photo sharing legal?
It depends on where the event happens and how the data is handled. GDPR Article 9 prohibits processing biometric data to identify someone unless an exception applies, and explicit consent is one. Illinois, Texas and Washington each have biometric laws with notice, consent, retention or destruction rules. This is general information, not legal advice.
What should a guest do before taking a selfie?
Read the notice on the selfie screen, check what the selfie is used for, and find out how to ask for deletion. Take the selfie in even light, facing the camera, with sunglasses and masks removed.

Sources

Try it at your next event

2,000 photos a year on the Free plan. No card.

Get Started

Free plan · No credit card · 5-minute setup

One QR. Every Guest.
Photos Delivered Live.

Start free with 2,000 photos. The event’s still on, and the photos are already theirs.

Early access

Join the ShotKite waitlist

We’re opening access in batches. Get an invite first.