Identification
How to Identify Any LEGO® Minifigure From a Photo — No Name, No Number
You are holding a figure you cannot name, and searching "blue knight with gold crown" gets you nowhere. Here is how a photo becomes an identification — and how to take a photo that works first time.
The short answer
Identification runs in two stages. A detector first finds and crops each figure in the frame; an identifier then matches each crop against a reference library and returns a name with a confidence score. Photograph figures upright, front-facing and not overlapping, on a plain background in soft daylight — light and angle change the result far more than camera quality does.
Key takeaways
- Identification happens in two stages, not one: a detector finds each figure in the frame, then an identifier names each crop.
- Light and angle matter more than camera quality. A modern phone is plenty; a shadow across a torso print is not.
- Front-facing, upright, and not overlapping is worth more than any setting you can change.
- When a scan won't commit, search the catalogue of 19,041 minifigures by name instead of accepting a guess.
- Every figure you confirm makes the next scan better — corrections feed straight back into the model.
Text search fails for minifigures, and it fails for a specific reason: you can't describe what you're holding in the words the catalogue uses. You see "a blue knight with a gold crown". The catalogue calls it something like cas315, filed under a theme you have never heard of. There is no query that bridges that gap.
A photograph does bridge it. Here is what happens between the shutter and the answer.
Stage one: finding the figures
The first surprise is that naming the figure is the second step. Before anything can be identified, something has to work out where the figures are in your image.
That is a separate model with a single job: draw a box around each minifigure in the frame. It is not trying to recognise anybody. It is answering "how many figures are in this picture, and where is each one".
This is harder than it sounds on a real photo:
- Figures overlap and hide each other.
- Some face away, some lie down, some are half-buried.
- Plenty of things in a LEGO photo are nearly figure-shaped — a loose torso, a brick-built animal, a flat black tile at low resolution.
Brick Loupe trains its own detector on its own photographs of real piles, and the property it is tuned hardest for is not inventing figures that were never there. In a scanning tool, a phantom detection is much worse than a missed one: you can rescan for a miss, but an imaginary figure quietly pollutes your inventory and your valuation.
Each box the detector finds is cut out as its own small image — a crop. Those crops, not your original photo, are what gets identified.
Stage two: naming each crop
Now each crop goes to the identifier, which compares it against a reference library of known minifigures and returns the closest match, along with how confident it is.
The important design decision here is what happens at low confidence. Brick Loupe's identifier is built to commit or abstain — it either returns a match it believes in, or it returns nothing. It does not pad the list with plausible-looking near-misses.
That is the honest behaviour, and it is also the less flattering one. A tool that always produces an answer looks more capable in a demo and is less useful in practice, because you cannot distinguish its good answers from its bad ones after the fact.
When a result comes back weak, you get a "Maybe?" prompt that re-runs identification on that crop. Often a second attempt with a different angle settles it.
Stage three: the price
Once a figure has an identity, it has a market value. Brick Loupe looks that up from recent sales data — what figures have actually sold for, not what sellers are asking — and shows it per figure, in your currency, converted at daily reference rates across 14 supported currencies.
Results stream as they complete, so a photo of forty figures starts filling your screen immediately rather than making you wait for the whole batch.
How to photograph figures so the scan works first time
Nearly every disappointing scan is a photograph problem, not a model problem. Five minutes of setup changes the hit rate dramatically.
Light
- Diffuse daylight is the best option available to you. A window on an overcast day is ideal. Direct sun creates hard shadows across torso printing, which is exactly the detail identification depends on.
- Avoid single overhead spotlights. They put each figure's face and torso in its own shadow.
- Don't use the flash if you can avoid it. It flattens printing and blows out the glossy plastic.
Background
- Plain and contrasting. A light-coloured towel, a sheet of paper, a plain table. Light figures on a dark ground, dark figures on a light one.
- Not a LEGO baseplate, not a carpet. Busy backgrounds full of brick-shaped texture make the detector work harder for no reason.
Arrangement
This is where most of the gain is:
- Stand figures upright if you have the patience. A standing figure shows its face and torso — the two most identifying surfaces.
- Front-facing. A figure photographed from behind is a plain back of a head and a torso, and often genuinely unidentifiable.
- Don't overlap. Leave a small gap between figures. Two figures touching can read as one box.
- Fill the frame, but not too much. Twenty to forty figures in one photo works well. Two hundred in one shot means each figure is a handful of pixels — split it into batches.
- Hold the camera parallel to the table. Shooting at a sharp angle foreshortens every figure.
Focus
Tap to focus on the middle of the group and hold the phone steady. Blur affects printed detail before it affects anything else, and printed detail is the whole game.
One tip worth more than the rest: if you have a handful of figures you particularly care about, photograph them separately, two or three to a frame, upright and front-on. Get the bulk of the box in wide shots and give the interesting figures their own portrait.
What to do when it still won't identify
Some figures cannot be identified from a photograph, by any tool. Usually one of these is why:
- It is a parts mash-up, not a catalogued figure. Somebody built it from three different figures. There is no correct answer because the thing does not exist in any catalogue.
- It differs from a common figure by one printed detail that is not visible in your shot.
- It is a reproduction of a rare figure, which is worth noticing for other reasons.
- It is genuinely obscure — a regional promotional item or a short-run exclusive outside the reference library.
For all of these, the route forward is the same: search the catalogue. Brick Loupe carries 19,041 minifigures and 24,888 sets, searchable by name, and a manual add lets you set quantity and condition and drop the figure into your collection alongside your scanned ones. A hand-added figure is priced exactly like a scanned one.
Your corrections make it better
When you confirm an identification — or fix a wrong one — that confirmation is kept as a labelled example. Those labels are what the identification model trains on.
This is not a courtesy feature. It is how the tool improves: every confirmed figure is one more real-world example of what that figure looks like in a real photograph under real light, which is exactly the data that a reference library of clean catalogue renders does not contain. Correcting a bad identification is the single most useful thing you can do, and it takes one tap.
If you would rather none of that happened, you can delete your account and your collection at any time, and the privacy policy spells out what is kept and what isn't.
Try it on the figure in your hand
Scan a photo in your browser — nothing to install — or get the app and point the camera at it. Free to try, no account needed to start.
Frequently asked questions
How do I identify a LEGO minifigure I can't name?
Photograph it and scan the photo. Brick Loupe locates each figure in the image, matches it against a reference library, and returns the identification plus its current market value. You need no name, number or theme to start — the photo carries all the information.
What is this minifigure? Can an app tell me from a picture?
Yes, in most cases. Point a phone camera at the figure in diffuse light, front-facing and upright, against a plain contrasting background. Brick Loupe returns the figure's identity and value. Where it cannot commit to an answer it says so, and you can search a catalogue of 19,041 minifigures by name instead.
Why did the scan miss some figures in my photo?
Almost always light, angle or overlap. Figures facing away, lying under others, or lit by a single harsh overhead lamp are the common causes. Spread the figures with small gaps between them, stand them upright and front-on, shoot in soft daylight with the camera parallel to the table, and rescan.
How many minifigures can one photo contain?
Brick Loupe is built for dense photos and handles dozens of figures in one frame. Twenty to forty works very comfortably. Past that, each figure occupies too few pixels for its printing to be readable, so split a very large collection into several photos rather than one enormous one.
Does it work on photos I didn't take myself?
Yes. You can drop in any image file in the browser scanner, and the Chrome and Firefox extensions let you click a minifigure photo on any web page — including a marketplace listing you are deciding whether to bid on — and scan it in place.
What if the figure is a mix of parts from different minifigures?
Then no catalogue contains it, and no identifier can name it correctly. Identify the individual parts if you care about the value, or add the closest catalogued figure by hand from the search. Mash-ups are generally worth the sum of their parts rather than a figure price.
Scan your minifigures now
Point your camera at a pile of figures, or drop a photo into the browser. Every figure gets identified and priced — free to try, no account needed for your first scans.