Recently, Nano Banana has been explosively popular, with many novel use cases emerging online, such as generating figurines, image restoration, and architectural image generation models. Attracted by these, we also tried it out and discovered two small applications related to GIS. Today, we'd like to share the prompts with you (Code is Cheap, show me the talk).

Scenario 1: Inferring Geographic Location from a Photo

This is a fascinating field. Many of you have probably seen videos where bloggers successfully deduce the shooting location of a photo by combining various factors like maps, culture, lighting, and geology—sometimes even pinpointing the exact building. Watching these videos gets us excited, but despite several attempts, we often find ourselves stuck without a clue. However, with AI like Nano Banana, the barrier to this functionality has been significantly lowered.

Prompt:

Infer the shooting location of the image

Original image:

Execution result:

This photo was taken in New York City, USA. Inference basis:

  • Store sign: The clearly visible "BLOOMINGDALE'S" sign on the left building is a famous American upscale department store with a flagship location in New York City.
  • Taxi: The yellow taxi in the photo is New York City's iconic "Yellow Cab."
  • Architectural style: The surrounding skyscrapers and urban street style are highly consistent with New York City's landscape.
  • Flag: The American flag hanging on the building further confirms the location.

However, after testing, we found that this result is only reliable for photos taken in places like the USA. For photos taken domestically, it mostly makes blind guesses. After careful testing with domestically taken images, domestic AI performs more stably. For example, this image:

Nano Banana inferred it was taken in Shanghai, but it was actually shot at Wuchang River Beach. Using Yuanbao's model, the result was:

Based on available information, I infer that the most likely shooting location is Wuhang (e.g., shooting the Hankou skyline from Wuchang River Beach) or Shanghai (e.g., shooting the Pudong skyline from near the Yangtze River Estuary).

Reasoning illustration:

We also tested other images, such as Xianning 131 and Lulin Lake in Lushan, and found that domestic DeepSeek significantly outperformed Nano Banana, providing thorough reasoning results.

Scenario 2: Generating Photos Based on Maps

This is currently a popular use case. Simply provide a map, draw an arrow, and let the AI generate a photo. In this scenario, the Nano Banana model performs exceptionally well.

Prompt:

draw what the red arrow sees

Input and output result:

However, after testing, we found that this result is only reliable for the USA or relatively large domestic cities. The test results for Shanghai were quite good, but using a map of Wuhan generated mostly composite photos. We speculate this might be due to insufficient street view data collection.

Scenario 3: Generating Photos Based on Topographic Maps

Those in GIS must remember the pain of being dominated by densely packed contour lines. Many experts can roughly discern the terrain just by looking at a contour map, but non-professionals might struggle. With Nano Banana, this scenario can also be easily implemented.

The prompt is the same as above:

draw what the red arrow sees

Input and output result:

Isn't it amazing? It's pretty much what we imagined. However, note: Do not upload classified maps! Do not upload classified maps! Do not upload classified maps!!!

Conclusion

These are the applications we've discovered so far. If there are other better directions, feel free to share them in the comments. Also, please follow Malagis for more updates. If there are other use cases, we will update them as soon as possible—after all, maps are images too～