A free daily geography game. Everyone in the world gets the same AI-generated outdoor scene each UTC day and has five guesses to name the country it depicts, with a distance, a direction and a proximity colour after each guess.

Where do the pictures come from?

They are generated by an AI image model from a written description of a real place — vegetation, terrain, building materials, which side of the road traffic uses, the writing system on signage, the angle of the light. They are not photographs, not street-view imagery, and never a specific building or landmark.

How do you know the answer is right?

Because the answer exists before the picture does. A place is picked from a hand-curated list, a description of its visible geography is written from that place, and the image is generated to match the description. The answer is therefore given by construction rather than inferred from the picture afterwards. A person then checks every single puzzle for contradictions between the scene and the answer before it can be scheduled.

Does an AI choose the places?

No. Locations come from a list a person curates by hand, which excludes disputed territories, active war zones and religiously sensitive specific sites. The generation step is not allowed to invent a location that is not on the list.

How the daily puzzles are made

Every GuessGlobe puzzle starts from a real place chosen off a hand-curated list; a description of that place's visible geography is written first and the scene is generated to match it, so the answer is fixed by construction rather than guessed from the picture afterwards.

Source: GuessGlobe production notesVerified 2026-08-04

The order matters more than the technology

There are two ways to build a game like this, and only one of them produces answers you can argue with.

What we do: pick a place, write down what is actually visible there — the vegetation band, the terrain, what roofs are made of, which side of the road traffic drives on, which writing system appears on signs, how high the sun gets — and then generate a picture that satisfies that description. The answer key is the description. It existed before the image did.

What we do not do: generate an attractive picture and then ask a model where it looks like. That produces an answer nobody can check, including us, and it turns the game into a guess about a model's opinion rather than about geography.

How the pictures are made, and why none of them is a photograph

Less of this is automated than you might assume, and the automated part is the picture. The place, the description of what is visible there, and all five clues are written by hand as rows in a curated list —Locations come from a list a person curates by hand, below, is the same list. Nothing picks a country, and nothing writes a clue, but a person.

What is automated is the step between: that description is concatenated into a single instruction and handed to an image generator unchanged. The parts of the instruction that never vary are:

A photorealistic wide establishing shot of an ordinary outdoor scene.
No readable text, no signage lettering, no watermarks, no logos, no flags of any kind, no people’s faces.
No famous landmark and no identifiable building; an unremarkable everyday view.

Between those sits the day's own description — vegetation, terrain, architecture, roads, environmental lettering, light — and each archived day publishes exactly that table, so you can read what the picture was asked for.

Every one of the 51 days published so far carries a generated scene. None of them is a stand-in any more. The archive was drawn with a local placeholder until 2026-08-25, and those days were rewritten rather than left that way: each stood on a whitelist row that named the thing people travel to photograph, so the picture came back as the postcard, and nobody had played them.

What a person checks, and what they do not

Every puzzle, without exception, is checked for the answer key being right and for contradictions between the scene and the answer — a country that drives on the left drawn with right-hand traffic, a tropical answer drawn with conifers. A puzzle that fails is thrown away and regenerated; it is never edited into shape.
About a third of them get a second look for image quality, composition and difficulty. That is a sampling rate, and we say so rather than implying every picture was art-directed.
Nobody edits a puzzle. The reviewer's only powers are to accept or to reject. An editable puzzle is a puzzle whose answer key can drift away from its picture.

What is deliberately not here

No street imagery and no map tiles from a street-view provider. Not as a claim of superiority — several excellent games are built on exactly that — but because this site is not built on anyone else's photographs.
No readable text in any scene. Signage may suggest a writing system, but a legible word would either give the answer away or be nonsense.
No landmarks. The scenes are ordinary views. A recognisable monument would make the puzzle a memory test.
No claim of being first. AI-generated imagery in geography games is not new: GeoGuessr hosts an AI Generated Quiz on its own site and other independent products use generated images too. Generated scenes are a choice about where pictures come from, not a novelty claim.
No player statistics on the archive. Scoring happens entirely in your browser and no puzzle number, answer, solved state or guess count reaches the analytics stream, so it cannot produce archive participation or solve-rate statistics. Publishing invented numbers would be worse than publishing none.

When we get one wrong

Sooner or later a puzzle will be wrong — a wrong answer key, or a scene that contradicts its own answer. When it happens, the plan is fixed in advance: mark the correction on the puzzle's page and its archive entry, replace the day's puzzle if the day is still running, and re-check every other puzzle made the same way. A daily game's mistakes are public by definition, so the correction has to be public too.

Questions about how it is made

What is GuessGlobe?: A free daily geography game. Everyone in the world gets the same AI-generated outdoor scene each UTC day and has five guesses to name the country it depicts, with a distance, a direction and a proximity colour after each guess.
Where do the pictures come from?: They are generated by an AI image model from a written description of a real place — vegetation, terrain, building materials, which side of the road traffic uses, the writing system on signage, the angle of the light. They are not photographs, not street-view imagery, and never a specific building or landmark.
How do you know the answer is right?: Because the answer exists before the picture does. A place is picked from a hand-curated list, a description of its visible geography is written from that place, and the image is generated to match the description. The answer is therefore given by construction rather than inferred from the picture afterwards. A person then checks every single puzzle for contradictions between the scene and the answer before it can be scheduled.
Does an AI choose the places?: No. Locations come from a list a person curates by hand, which excludes disputed territories, active war zones and religiously sensitive specific sites. The generation step is not allowed to invent a location that is not on the list.

Play today's puzzleOne scene, five guesses, the same puzzle everyone else gets today.