This article is designed as a practical prompt-training exercise for architects and design students exploring AI workflows. The approach is intentionally simple: begin with a single, well-defined project image, then use Gemini 3.1 Pro to systematically generate the remaining views — one at a time. Instead of relying on randomness, the process focuses on controlled iteration, where each output builds on a clear visual foundation. Effective prompting doesn’t require complexity.

In this article, we explore Stacked Horizon , an experimental architectural concept generated using NanoBanana 2 . The project demonstrates how structured prompts can guide AI to develop architectural massing, façade articulation, and visual presentation. Below, we share the original base image followed by the generated variations along with the exact prompts used to create them. The Base Image Every workflow begins with a base reference image. This can be a rough massing m