What are common challenges in deploying generative models for real-time applications?