You can start out by looking at deploying stable diffusion models with accelerated inference (https://github.com/stochasticai/x-stable-diffusion). You normally want to be containerizing the models when you deploy them, easier to maintain and scale. DM me if you have any questions.