Quick-Start Guide: Getting the Most Accurate Guess From Raven

An AI vision model can only work with what's actually visible in a photo — so the fastest way to get a sharper guess out of Raven isn't a setting or a trick, it's just handing it a photo that's giving it something real to look at. Here's a quick checklist to run through before you hit upload, whether you're testing it for fun or genuinely curious about a photo you've had lying around.

Before You Upload

1. 1.Pick a photo with something in the background. A landmark, a road sign, a shopfront, even a distinctive plant or roofline gives the model far more to work with than a tight close-up of a person or object with nothing around it.
2. 2.Use the highest resolution version you have. A compressed, blurry copy hides exactly the kind of fine detail — small text, tile patterns, distant signage — that separates a strong guess from a vague one.
3. 3.Skip the heavy filters. Aggressive color grading, black-and-white conversions, and stylized filters can wash out cues like foliage color and sky tone that genuinely factor into a location guess.
4. 4.Leave crops generous. Cropping in tight to "help" the AI usually backfires — it just removes the surrounding context that would have narrowed things down.
5. 5.Shoot in daylight when you can. Even lighting makes architectural and vegetation details easier to read; harsh shadows or very low light can obscure details a clearer shot would reveal.

What to Avoid

A few habits reliably work against you. Screenshotting a photo instead of uploading the original file adds a second layer of compression on top of whatever the photo already had, quietly degrading the exact fine detail the model relies on. Photos with a watermark or a UI overlay across the middle of the frame can obscure the very clue that would've made the difference. And photos that have been mirrored or rotated for a social post occasionally throw off details like text orientation — if you have the original, unflipped version, that's the one worth using. It's also worth resisting the temptation to crop out people in the frame if doing so means cutting away the background behind them — a person can be cropped later; the storefront over their shoulder usually can't be recovered once it's gone.

If You're Choosing From a Batch

When you've got several candidate photos of the same trip or the same mystery location, it's worth being a little deliberate about which one you upload first. Pick the frame with the most going on around the edges rather than the best-composed one — a slightly awkward shot that happens to catch a street sign or a distinctive railing in the corner will usually outperform a beautifully centered photo that crops all of that out. If the first result comes back vague, try a second photo from the same set before assuming the location just isn't guessable; different frames from the same moment often carry different clues.

Reading the Result

Once you get a guess back, the confidence score is worth paying attention to, not just the named location. A high-confidence result usually means the photo had a cluster of strong, specific clues — legible signage, a recognizable building style, a distinctive landscape. A lower-confidence result isn't a failure; it's the model being honest that the scene was more generic and could plausibly be a few different places. Either way, treat the guess as exactly what it is: a best read based on visual evidence, offered for fun, not a verified fact.

Testing on Location

If you want to push this further, Geospy AI — our sibling app, available on the App Store — is worth trying the next time you're actually somewhere interesting. Since it's built for your phone, you can snap a fresh photo on the spot and see in real time how much a clean, well-lit, uncropped shot changes the quality of the guess compared to an old, low-resolution photo pulled from a group chat. It's a genuinely fun way to feel out, firsthand, which of these tips actually move the needle.

The Short Version

Give it a real photo with background detail intact, keep the resolution and framing generous, skip the filters, and treat the result as a fun, well-reasoned guess rather than a verdict. Head to withraven.net with a photo you're curious about and see what it notices — that's really the whole quick-start guide in one sentence.