🗺 ChatGPT Can Identify Locations from Photos
The new OpenAI models o3 and o4-mini reason not just with text but also with images. Users have discovered that this allows the models to pinpoint cities, neighborhoods, and exact locations based on the tiniest visual clues in a photo.
📸 How does it work?
The model doesn't rely on metadata or chat history — it concludes purely from visual details: architecture, signs, language, and trees ⤴️ It can rotate and zoom in on images if needed.
To try it yourself, upload a photo to ChatGPT and use this prompt:
You are participating in a geolocation challenge. Based on the provided image:
1. Carefully analyze the image for clues about its location (architecture, signage, vegetation, terrain, etc.)
2. Think step-by-step about what country this is likely to be in and why
3. Estimate the approximate latitude and longitude based on your analysis
Take your time to reason through the evidence. Your final answer MUST include these three lines somewhere in your response:
country: [country name]
lat: [latitude as a decimal number]
lng: [longitude as a decimal number]
You can provide additional reasoning or explanation, but these three specific lines MUST be included.
The model can return the location coordinates where a photo was taken. Even simpler models like GPT-4o often succeed for well-known landmarks or streets in major European or U.S. cities. However, the more complex the location, the more apparent the new models' "visual reasoning" advantages will be.
🛡 OpenAI claims safety is a priority: the model is supposed to "decline requests to identify faces or private information."
