The Best (and Worst) Photos to Upload for an Accurate Guess

Not every photo gives an AI model the same amount to work with, and that's true whether you're testing Raven for fun or just curious how it reasons. Some photos are stacked with independent, corroborating clues; others are almost deliberately unhelpful, cropped down to the one detail that tells you the least. Knowing the difference makes the whole experience more interesting, and it also just means fewer vague, unsatisfying guesses.

None of this is about gaming the system — Raven isn't grading you, and a vague guess on a hard photo isn't a failure. But if you're curious what actually gives a model the best shot at a specific, confident answer, there's a pretty clear pattern.

What Gives the Model the Most to Work With

The best photos, without exception, are the ones that show more than one thing at once. A wide shot of a street, a hillside, or a town square captures architecture, vegetation, sky, and often some signage or infrastructure all in the same frame — which means several independent clues can corroborate each other instead of the model having to guess from one detail alone. Daylight helps too, simply because natural light renders color, texture, and shadow more legibly than a dim interior or a heavily backlit sunset silhouette. And a photo that hasn't been aggressively cropped, filtered, or stylized preserves more of the original visual information — the same information a heavy filter or crop tends to throw away first.

• —Wide environmental shots — a street, a hillside, a shoreline — that show architecture, terrain, or vegetation alongside the sky.
• —Natural daylight, which renders true color and legible shadow direction far better than artificial or low light.
• —Unedited or lightly edited images, since heavy filters and color grading can shift the exact tones — soil color, sky color, foliage color — the model is reading.
• —A visible horizon or some sense of scale, which helps distinguish, say, distant mountains from a nearby rock formation.

The Photos That Trip Up Any Model — Human or AI

On the other end, a handful of photo types reliably produce vaguer, more hedged guesses, and it's not really a limitation specific to any one model — a person would struggle with these too. Extreme close-ups are the biggest culprit: a macro shot of a single flower, a plate of food, or a stretch of pavement with nothing else in frame removes almost every environmental clue at once. Heavily filtered or stylized photos — strong color grades, black-and-white conversions, or aggressive cropping down to a single subject — strip out exactly the color and context information that clues like soil, sky, and foliage rely on. And generic indoor scenes are their own category of hard: hotel rooms, chain coffee shops, and big-box retail interiors are deliberately designed to look similar everywhere, which is great for brand consistency and terrible for geolocation.

A Few Composition Habits Worth Building

If you're specifically trying to get a confident, specific guess rather than a broad regional one, it helps to think like you're gathering evidence rather than just taking a nice photo. Include a little more of the scene than feels necessary — a bit of sky, a bit of ground, whatever's around the edges of your main subject. If there's a sign, a license plate, or any text visible, don't crop it out just because it's not the focal point; it's often doing more work than the subject itself. And if you're choosing between two similar shots, pick the one with more variety in it — a street with both a building and some trees beats a street with just a wall, even if the wall photo is the better composition.

A Vague Guess Is Still a Real Answer

It's worth saying plainly: some photos are just hard, and that's fine. A close-up of a rock or a generic hotel hallway might only get you a guess at a broad climate zone, or an honest "could be several regions" — that's a fair reflection of how little the image actually gives away, not a flaw in the analysis. The fun of it is genuinely in the range, from a photo so specific it nails a city to one so sparse it can only offer a continent.

Next time you're picking a photo to test, try uploading two versions of the same trip — a wide shot and a tight crop — to Raven at withraven.net and see how differently Gemini talks about each one. If your best travel photos are already sitting in your phone's camera roll, the sibling app Geospy AI does the same kind of analysis on iOS, available on the App Store, so you can run the same experiment without leaving your gallery. Either way, the wider the shot, the better the story it can tell.