Architectural Prompting: modeling in a prompt, AI and Landscape Architecture

The test

It all started with a simple screenshot from Google Maps. As often happens, I imagined myself in the preliminary phase of a project to be developed in a rural area. Instead of spending hours tracing a map or building the model from scratch, I decided to rely on Dwarf Banana.

I uploaded the screenshot to NanoBanana asking: “Generate a CAD drawing-style technical floor plan from this satellite view.”

The result was very interesting. The plan shows very faithfully the streets, the built lots, the trees, even the agricultural boundaries. Obviously it’s not perfect – some lines don’t exactly follow reality – but considering it’s top-level output, achieved in less than 30 seconds, the potential is enormous.

I made a simple overlap between the original image and the generated floor plan, to check the fidelity. The result? The AI ​​correctly interpreted the main geometries and proportions, reconstructing a coherent system.

At this point I asked NanoBanana: “View this technical floor plan as a wooden architectural model, competition plastic style, low perspective view.”

And here is a scale model that we could imagine on a review table: stylized buildings, land relief, trees and crops. Some buildings are “invented” (the AI ​​does not have real volumetric data), but the overall coherence remains extraordinary.

Still on the same layout, I asked: “Keep the layout but change the style of the materials, making it clearer and more minimal, with a higher room.”

The result is an elegant, clean model with a more Nordic and competition look. Useful for a general representation.

Last request: “Provide a close-up detail of the architectural model, with focus on vegetation, road and buildings in the foreground.”

And here is an immersive close-up: very distinct materials, human presence, vegetation modulated with great taste.

What is striking is how little it took to get all this: a screenshot from Google Maps and a few well-formulated requests. In just a few minutes, the machine constructed a complete sequence – technical plan, architectural model, and immersive detail – starting from very limited data.
Of course, there are still margins of error: some proportions may not be perfect, some volumes are interpretations. But let’s remember that we are talking about a model launched just over two months ago. And the prospects, for those who work in the architectural or landscape field, are truly exciting.

Conclusions

Dwarf Banana it is not (yet) a precision instrument. But it is a powerful previsualization tool: in a few minutes you can obtain a clear territorial layout, transform it into a model, test views and presentation approaches. For architects, urban planners and designers, this is a new grammar of the project, where the speed of visualization can really make the difference in the concept phase or in dialogue with the client. And as always, we’re just the beginning! See you soon with new tests!