Building upon the foundational deployment strategies covered previously, this chapter addresses more complex operational requirements often encountered when running diffusion models at significant scale. You will learn techniques for distributing models geographically to reduce latency and improve availability through multi-region architectures. We will cover methods for safely introducing changes, such as A/B testing new model versions or sampling parameters. Furthermore, this chapter presents advanced strategies for cost optimization, including the effective use of spot instances and ways to mitigate their interruptions. Finally, we will examine specific operational challenges like managing cold start latencies and configuring load balancing suitable for the long processing times characteristic of diffusion model inference.