Skip to content
Raven

§ Method comparison

Find a Location From an Image

There are only three honest ways to work out where a picture was taken, and they answer completely different questions. One reads data the camera left behind. One looks for the same view somewhere it has already been published. One reasons from what is visible in the frame. Knowing which to reach for first saves far more time than any single tool does.

This page compares all three plainly — including where the one Raven uses gives up. It is not a ranking; the right route depends entirely on how the file reached you and what is in it.

§ The three routes

What each one is actually doing.

Read these in order. They are listed the way you should try them, not the way they are usually marketed.

  1. 01Exif

    Read the file's own metadata

    Cameras and phones write a small data block into the image file itself — the timestamp, the camera model, and, if location services were switched on, the coordinates the shutter fired at. When that block survives, nothing is being deduced. You are simply reading what the camera already recorded.

    Strongest when
    Files that reached you byte-for-byte: straight off a memory card, AirDropped, emailed as an attachment, or pulled from a cloud drive.
    Where it breaks
    Most platforms re-encode what you upload and drop the location block on the way in, so a picture that arrived through a social feed or a chat app usually carries nothing. A screenshot never had the original block to begin with.
    What you get back
    Exact coordinates, or nothing at all. There is no halfway answer, and no tool can recover a block that was stripped.
  2. 02Reverse search

    Match the image against published pictures

    Search engines index enormous numbers of photographs. Reverse image matching compares your file against that index and looks for the same frame — or something close enough — that somebody already published with a caption, a place name, or a map pin attached to it.

    Strongest when
    Anything that has been photographed for the internet before: landmarks, hotel rooms in listings, restaurant interiors, stock photography, artwork, packaging, and the view from a well-trodden viewpoint.
    Where it breaks
    An ordinary residential street that nobody had a reason to publish. The search still returns results, but they are visual look-alikes rather than the same place, which is easy to mistake for a hit.
    What you get back
    Either a near-certain match you can open and read for yourself, or a page of similar-looking pictures that prove nothing.
  3. 03What Raven does

    Reason from what is visible in the frame

    A vision model looks at the picture the way a widely-travelled person might, and works from what is actually in shot: the script on a shopfront, the shape of a kerb, the colour of a number plate, the plants along the verge, the angle of the light. It never opens the file's metadata.

    Strongest when
    Ordinary scenes with no coordinates left in the file and no published twin online — precisely the photographs the first two routes go quiet on.
    Where it breaks
    Blank interiors, tight close-ups, night shots, and regions whose architecture and signage are shared across several countries. It degrades into a broader guess rather than refusing to answer.
    What you get back
    Always something: a country, region and city estimate, a confidence score, and a short account of which details led there. Often the right region, frequently the wrong city, sometimes simply wrong.
Side by side03 ROUTES
MethodBest caseReturns nothing whenWhat it costs you
Metadata (Exif)Exact coordinates — a reading, not an estimateThe file passed through a platform that stripped it, or it is a screenshotSeconds, no tool needed — your operating system already shows it
Reverse image matchingExact place, when the same subject has been published beforeThe subject is unremarkable and has never been indexedA minute, plus judgement to tell a real match from a look-alike
Visual reasoningRegion-level estimate, occasionally street-levelNever silent — it answers even when it should not be confidentOne upload; the reasoning comes back with the guess

The two comparisons worth reading before you commit to a route: reverse image search versus AI geolocation and, for the tool most people already have open, how Google Lens differs from AI geolocation.

§ Route three

Try the visual route now.

If you have already checked the metadata and already tried matching, this is the route that is left — and the one that always returns something. Drop the photo in and read the reasoning as carefully as the place name.

  • The file's metadata is never opened. Raven works from the pixels alone, so a stripped photo and an untouched one get exactly the same treatment.
  • The image is held in memory while the model looks at it and discarded when the answer comes back. It is not written to a disk, a bucket or a database.
  • You get a confidence score and the details that drove the guess — which is what makes the result checkable instead of just assertive.

Curious which details it leans on? They are catalogued in 10 visual clues AI uses to find a location rather than repeated here.

Sign in to start guessing

Raven needs a quick sign-in before your first photo. It's free to start — no card required to try.

§ Decision guide

In what order, and when to stop.

Cheapest and most exact first, vaguest last. Each step is worth doing only because the one before it came back empty.

  1. 01

    Open the file and look for coordinates

    It costs nothing and it is the only route that can be exactly right. On a Mac, open the image in Preview and choose Tools, then Show Inspector, then the GPS tab. On Windows, right-click the file, open Properties, and read the Details tab. On an iPhone, open the photo and swipe up — a map appears if coordinates are attached. No GPS section means the data is gone.

  2. 02

    Crop to the most published-looking thing in frame

    If the metadata is empty, reverse image matching is the next cheapest attempt — but it rewards a tight crop. A whole street rarely matches anything; a signposted building, a statue, a distinctive shopfront or a hotel interior stands a far better chance of having been photographed and captioned by somebody else.

  3. 03

    Hand the frame over for visual reasoning

    When the file is bare and the matching returns look-alikes, reasoning from what is visible is the only route left. Treat what comes back as a hypothesis with a region attached, not as an address — and read the explanation it gives you more carefully than the place name.

  4. 04

    Confirm the hypothesis somewhere else

    Whatever route produced a candidate, the candidate is worth as much as the check you run on it. Take the suggested city and compare street-level imagery, sun angle, and the details in shot against the real place before you believe any of it.

Step four is the one people skip, and it is the one that decides whether any of this was worth doing — how to verify where a photo was taken walks through the checks. For step one, how to remove location data from photos shows you where the Exif block lives on every platform — useful for reading it as well as for stripping it.

§ Limitations

What none of these three can do.

Metadata cannot be recovered once it has been stripped. No amount of processing puts coordinates back into a file that no longer carries them, and any tool claiming otherwise is guessing while calling it a reading.

Matching cannot find what was never published. A quiet street with no reason to appear online will keep returning look-alikes no matter how many services you try it on.

Visual reasoning cannot decline to answer. That is the trap in the third route: it produces a plausible-sounding place for every photograph, including the ones where the frame simply does not contain enough to go on. The confidence score is the model's own impression of the evidence, not a measured probability — read it as a hint about how much the picture had to offer, and nothing more.

And none of the three can tell you who is in a photograph or where a person is. Raven is an entertainment tool for curiosity about places; it is not a way to find people, and it should never be pointed at that problem.

Important notice

Raven returns AI-generated estimates, not verified facts. Guesses are approximate and can be completely wrong. Raven is built purely for entertainment and curiosity — it is not a surveillance tool and must never be used to track, locate, or identify real people. Raven is an independent product and is not affiliated with, endorsed by, or connected to any government agency or law-enforcement body.

§ FAQ

Choosing between the methods.

Which method should I try first?
Metadata, every time. It is the only one of the three that can be exactly right rather than approximately right, it takes seconds, and it needs no tool beyond the file browser you already have. Move on to reverse image matching only once you have confirmed the coordinates are not there, and to visual reasoning only once matching has come back with look-alikes instead of the place itself.
Why does Raven refuse to read the photo's Exif data?
It is a deliberate design choice, not a missing feature. Raven works only from what is visible in the frame; the file's metadata is never opened, the image is held in memory while the model looks at it and discarded when the response is sent, and nothing is written to storage. It also matters less than people expect — most photographs that get shared have already had their location block removed in transit, so a metadata reader would return nothing on exactly the pictures people are curious about.
Reverse image search found nothing. Does that mean the photo is unusual?
No — it usually means the opposite. Matching can only find a place that somebody has already photographed and published, so it goes quiet on the most ordinary scenes: a residential street, a car park, a stretch of road between two towns. A blank result says the frame has not been indexed, not that the location is obscure or interesting.
Can the three methods be combined?
That is the most useful way to work. Use visual reasoning to produce a region-level hypothesis, then turn that hypothesis into a better search: once you have a plausible country or city, reverse image matching and mapping imagery have somewhere specific to look, which is far more productive than searching the whole world at once.
Is any of this accurate enough to rely on?
Only the metadata route, and only when the coordinates are genuinely present. Matching can be fooled by a similar-looking building, and visual reasoning returns an estimate that is often the right region and frequently the wrong city. Raven is built for entertainment and curiosity — treat every guess as a starting point for your own checking, never as an answer, and never as a way to locate a person.

Running the third route often? Geospy AI does it on iPhone without the one-guess limit.

Get Geospy AI — free →

§ Keep reading

The longer versions.

Each of these takes one part of the comparison above and goes properly into it.

Get Geospy AI for iPhoneDownload free