This research explores the automation of architectural visualizations, focusing on floor plans, using AI image generation technology. Traditionally, floor plans have been crucial for conveying design specifications to experts and non-experts alike. However, as architectural data volume and complexity increase, creating floor plans becomes time and resource intensive. Recent advancements in AI image generation, particularly Large Language Models (LLMs), have introduced innovative text-to-image techniques. This study presents a novel approach to design visualization by employing this technology for floor plan generation. First, a test assessed an image generation AI model, producing around 10,000 floor plan images. The need to generalize design knowledge and implement additional training methods for image generation AI was recognized. Second, design knowledge about floor plans was generalized. Technical training methods were categorized as 1) Data Preparation, 2) Hyperparameters Optimization, and 3) Training. The proposed approach was then demonstrated through floor plan generation scenarios. It provides a supportive tool for creating floor plan images that meet user requirements, allowing non-experts and clients with limited knowledge to visualize preliminary designs. The research also highlights the potential of image generation AI in various design visualization applications, including user-centered design and customization opportunities.