This study explores the challenges of generating multi-angle product images using diffusion models. We examined six products with varying structural complexities and found that a product’s ability to generate specific viewing angles is influenced more by its key design features and common display angles than by its complexity or centrality. Additionally, we assessed fine-tuning methods for generating safety glasses images. While results were not consistently stable, trained embeddings helped maintain the object’s identity across different angles. Our findings highlight the need for design-informed training strategies to enhance AI-generated product visualization.