Capturing
the effort,
before the memory.

In 30 seconds

Traditional methods and their limits

• ▸Sorting by hand : hours, sometimes days, spent reading bib numbers photo by photo.
• ▸The dump-it-all gallery : participants scroll through hundreds of pages to find their shots — half of them give up along the way.
• ▸Timing checkpoints : accurate, but limited to a few points on the course, and useless for sports without a timing chip.

AI solves all three limits in one move: it identifies, indexes and serves photos in near real time.

Three technologies come together in a modern sports photography platform: bib number detection with OCR, biometric face recognition, and an instant search engine. Here is how.

Real-time AI analysis

23 ms

3 sujets identifiés

What the AI sees

Faces detected

2— avg. confidence 95%

Bib numbers read

1— no. 703 — 99%

Analysis time

23 ms

Photo indexed

Associée aux 2 participantes et au dossard n° 703.

Bib number detection (adaptive OCR)

Classic OCR (optical character recognition) is not enough for sports photography. Bibs get twisted by the wind, creased by sweat, sometimes hidden behind an arm. Modern AI uses computer vision models trained specifically on this kind of image — it infers the number even when a digit is missing, using the surrounding context.

Biometric face recognition

The principle is simple to explain but technical to implement. Every face detected in a photo is turned into a vector with several hundred dimensions — its embedding. Two identical faces produce vectors that sit close together. When a participant uploads their selfie, the algorithm measures the distance between their embedding and every indexed one. The closest photos come back in under 100 ms.

The winning combo: OCR + face

Taken separately, each model has its limits. Bib hidden? OCR fails. Runner turning their head 90°? Face recognition fails. Combined, they cover more than 95% of cases at a typical event. The rare unidentified photos are surfaced in a manual review tool, where they get sorted within minutes.

This is probably the most misunderstood topic of all. Face recognition counts as biometric data under GDPR, which puts it in the category of especially sensitive data. But the regulation does not ban it — it frames it.

The 4 pillars of compliance

1. 01

   Explicit consent

   The participant has to actively accept the use of face recognition to search for their photos. A pre-ticked box is not enough. Bib number search always stays available as an alternative.

2. 02

   Limited purpose

   Biometric vectors serve exactly one purpose: helping people find their own photos. No advertising use, no resale, no cross-referencing with other databases.

3. 03

   Limited retention period

   Embeddings must not be kept any longer than necessary. Photag deletes them automatically at the end of the sales period.

4. 04

   Secure storage

   European hosting, encryption in transit and at rest, and immediate erasure on request.

Paradoxically, AI is more protective of privacy than an open public gallery. When an organizer posts 5,000 photos on Facebook, anyone can see every participant's face. With an AI platform, each participant only sees what concerns them.

The job of a sports photographer has long been ruled by one reality: post-production often took longer than the shoot itself. A day on a trail race means 6 hours in the field… then 12 hours sorting that evening and the next day. AI flips that ratio.

photag.fr/photographer

Live

The AI workflow in 3 steps

How the revenue adds up

Speed drives conversion directly. 5 times more sales when the photo lands within the hour, compared with the several-day wait of a manual workflow. At equal event size, a photographer equipped with AI regularly out-earns one working the classic way — not because they charge more, but because they sell to far more people.

For an organizer, photos are not a side service — they are a spin-off product with huge emotional impact. The memory of the finish line, of the forced smile at kilometer 32, of the last sprint out of the final bend. Give your participants those images right away, and you turn a sports event into a shared moment that lives on long after the race.

Three concrete levers

The opposite is just as true: a painful gallery (search by finishing time, a wait of several days, cut-rate quality) leaves a bad taste. It is often the last point of contact between an organizer and their participants — you may as well make it a good one.

Three families of solutions coexist on the sports photography market today. Picking the right one depends on volume, on how strict you are about GDPR, on how much time you want to spend on post-production, and on the business model you are aiming for.

Here are the criteria that really matter, compared across three typical approaches.

The questions to ask before you sign

• ▸Where are the photos and the biometric vectors hosted? (the answer should be in Europe)
• ▸What is the automatic purge policy for biometric embeddings?
• ▸Is the payment module built in, or do you have to plug one in yourself?
• ▸What commission is taken on each sale? Is there a fixed subscription on top?
• ▸Is there a manual review tool for the rare cases where the AI does not identify a participant?
• ▸Is the participant's mobile experience genuinely optimized, or just a desktop site hastily squeezed onto a phone?