10 Visual Clues AI Uses to Figure Out Where a Photo Was Taken

When Raven looks at a photo, it is not hunting for one giant, obvious clue like a famous landmark. Most photos do not have the Eiffel Tower in them. Instead the model weighs dozens of small, ordinary details at once — the kind of things a local would never notice, but that quietly narrow down a region and sometimes a country. What follows is the working list: the ten signals that carry the most weight, roughly in the order they tend to matter.

1. 1.Architecture and roofline. The pitch, material and shape of a roof is one of the strongest regional signals there is — steep metal roofs shed snow in the mountains, flat masonry roofs make sense where it barely rains, and terracotta tiles cluster around the Mediterranean and Latin America. Walls matter too: half-timbering, whitewashed render, vinyl siding and unrendered brick each belong to a different part of the map.
2. 2.Signage and script. Even an unreadable, blurry sign is useful. The model does not need to translate the words — recognising that the script is Thai, Cyrillic or Arabic instantly narrows the search space to a handful of countries, and the font style, sign shape and border colour add further texture on top of that.
3. 3.Vegetation and foliage. Palms, olive groves, birch stands and bamboo are each tied to specific climate bands. A single distinctive species in the background can rule out entire continents, and the state of the foliage — bare, budding, full, turning — narrows the season as well as the place.
4. 4.Road markings and lane paint. Solid yellow centre lines are common in North America and rare across most of Europe, where white dominates. Crossing style, kerb colour, the presence of a hard shoulder and the material of the road surface all vary by country in ways drivers never consciously register.
5. 5.Number plates and vehicle formats. The full registration is almost never legible, but the colour, proportions and border design usually are — a blue EU strip, a yellow rear plate, or the wider, squarer shape of a North American plate are all recognisable at a glance, even at low resolution.
6. 6.Sun angle and shadow length. Shadows encode latitude and season. Long, soft shadows suggest a high latitude or early morning; short, hard ones suggest somewhere closer to the equator near midday. The direction a shadow falls at noon even hints at which hemisphere the camera is in.
7. 7.Terrain and geology. Red laterite soil, black volcanic rock, pale karst limestone and glacier-carved valleys all have distinct visual signatures tied to specific parts of the world, and geology changes far more slowly than anything humans build on top of it.
8. 8.Utility infrastructure. Wooden power poles against concrete ones, the shape of a transformer box, the design stamped into a manhole cover, the style of a street light or a post box: unglamorous details that vary surprisingly consistently by country because they are procured nationally and replaced rarely.
9. 9.Car models and makes. Which manufacturers dominate the kerbside — and whether the steering wheel sits on the left or right — is a strong regional tell, especially combined with which side of the road traffic is actually using. Fleet age is a second layer: a street of twenty-year-old cars reads differently from a street of new ones.
10. 10.Coastal and water features. The colour and clarity of water, the shape of a coastline, the design of boats and jetties, and the presence of particular reef or mangrove systems can place a photo along a very specific stretch of coast, though open water alone is one of the hardest scenes there is.

No item on that list is a smoking gun on its own. What produces a confident guess is convergence: several weak signals stacking up in the same direction. Left-hand traffic, a yellow rear plate, red brick terraced houses and a flat grey sky together point strongly at the United Kingdom, even though any one of those clues alone would be far too vague to commit to. Around a third of the world's population drives on the left, so that clue alone still leaves dozens of countries in play — but it removes most of them, which is exactly the job.

Standardisation cuts both ways. The 1968 Vienna Convention on Road Signs and Signals pushed much of the world toward shared shapes and colours, which makes signage less discriminating in Europe than a newcomer expects. The countries that never adopted it, or adopted it partially, become correspondingly easier to spot. The same logic applies to climate: the Köppen system sorts the planet into five broad groups and about 30 sub-types, and vegetation is essentially a slow, visible read-out of which one you are standing in.

This is why some photos are far easier to place than others. A close-up of a plate of food gives almost nothing to work with beyond the cuisine, which travels. A wide street scene with a car, a sign, some greenery and a sliver of sky is a goldmine by comparison, simply because there are more independent clues available to line up against each other. The practical rules of thumb:

• —Include the ground and the sky. Road surface, kerb and lane paint at the bottom; light quality and cloud type at the top. Both are free clues most people crop out.
• —Keep one legible sign in frame, even a small one at the edge. Script is the sharpest single clue on the list.
• —Prefer a slightly wider shot over a beautifully composed close-up. Composition and evidence pull in opposite directions here.
• —Do not upscale or heavily filter. Recolouring changes soil tone, vegetation hue and sign colour, which are three separate clues at once.

Independence is the part most people miss. A terracotta roof, a whitewashed wall and an olive tree feel like three clues, but they are close to one clue counted three times: they all say Mediterranean climate. A terracotta roof, a number plate with a blue strip, a speed limit posted in kilometres and a sign in Greek script are four genuinely independent constraints, and their intersection is a great deal smaller. This is why a photo of an unremarkable street corner sometimes beats a photo of a spectacular view — the street corner mixes clue types, while the view repeats one.

Every clue above is learnable by eye, and each has its own field guide. Signage is the deepest well: start with road signs as geographic fingerprints for shape, colour and unit, then how script and language narrow down a location for the writing system itself. For the built environment, architecture styles around the world covers walls and plans, while the field guide to roofs and rooflines covers the single most visible surface in most outdoor photographs.

That layering is what makes this kind of analysis feel less like a lookup and more like detective work, and it is the whole idea behind Raven. Upload a photo at withraven.net and Google's Gemini weighs this same evidence in seconds; the step-by-step walkthrough shows what the result actually says and how to read it. The image is processed in memory for one request and never stored, the web tool gives one free guess per account, and the free Geospy AI iPhone app carries the same idea when you are away from a desk. Every answer is an estimate for entertainment, and it can be confidently wrong.