AI Architectural Removal and Distillation: A Path to Efficient Compression in AI Text-to-Image Diffusion Models