Blog Article
A starting point, not a final answer
How photo identification works in Reveliers, why I refuse to claim it nails the exact reference, and why the photos are not kept.
Point a camera at a watch and software will now tell you what it is. That sentence sells apps, and it is roughly one third true. I want to explain how identification actually works in Reveliers, because the honest version is still useful, and because the dishonest version is how you lose a collector's trust in one interaction.
What actually happens
When you snap a watch in Reveliers, a vision model reads the photo and extracts the attributes a person would: brand, model family, dial color, case material, bezel style, complications, a case diameter estimate. Those attributes then rank a catalog of tens of thousands of references spanning the brands collectors actually own, from Rolex and Omega to Grand Seiko, vintage pieces, and microbrands. You get back the most likely candidates, not a single proclamation.
From there the flow is deliberately manual. You pick the right reference, or the close one, and the record starts pre-filled instead of blank. Then you correct it and add what no photo can know: the serial, the purchase story, the box and papers situation, the service history.
Why not just claim the exact reference
Because watches make that claim impossible to keep, and collectors know it.
A date and a no-date variant can differ by one window. Dial variants of the same reference differ by a font. Two references can share a case and differ only in movement. Textured and reflective dials photograph as three different colors depending on the light. And a good homage is specifically engineered to look like the reference it honors. A one-shot "this is exactly reference 16610LV" answer will be wrong often enough to be embarrassing, and each wrong answer teaches you to distrust every other number in the app.
So the design goal is different: get you to the right starting point fast, and make correcting the details effortless. The app helps you build a better record. It does not pretend every watch can be perfectly identified from one image. When identification misses, that is a signal about the catalog or the photo, and there is a feedback path for exactly that.
I think of it the way I think of a knowledgeable friend at a get-together. "That looks like a mid-2000s Speedmaster, the reduced maybe?" is genuinely helpful. A friend who announced exact reference numbers with total confidence and was wrong a third of the time would be worse than no friend at all.
What happens to your photo
This part matters more than the model. Identify photos are processed to produce the match suggestions and they are not retained. Not stored for training, not filed away in a corner of a bucket. The photo does its job and is deleted. If you choose to post it to the community feed afterward, that is a separate action you take on purpose.
The same honesty applies to what identification is not connected to. Snapping a watch does not add it to your vault. Nothing enters your collection record unless you save it, and once saved, the private parts of that record are encrypted on your device like everything else in the vault.
The quiet payoff
The unglamorous truth is that identification is not the product. The record is the product. Identification exists so that starting a record takes seconds instead of minutes, because every skipped form field is a record that actually gets kept.
A starting point, not a final answer. It is a hedge, and I put it on the marketing site on purpose. In a hobby full of confident wrong answers, being reliably useful beats being impressively wrong.