Allow passing in lora from huggingface/civit.ai

Deploying a custom model works, but for sites with low traffic the cold-boot can be painfully long. 

I'm curious if we could add an optional parameter that'd point to a lora stored on huggingface (or civit.ai) that we can load and un-load after the generation. 

I'm happy to contribute if you give me a thumbs up. 