load insightface from extra_model_path

[xingren23]

https://github.com/cubiq/ComfyUI_InstantID/issues/37

and Import InsightFace as needed

[cubiq]

insightface is always needed for instantid so it can be loaded at startup

[xingren23]

Sure, that makes sense;
In my scenario, ComfyUI is utilized in a serverless fashion, where the ComfyUI container is launched for the duration of the execution and terminated post-execution. Since not every workflow necessitates the use of InstantIDFaceAnalysis, I am looking into ways to optimize the startup time.

[alexalok]

would love to see it merged!

[TrevorSquillario]

Would like to see this added as well. extra_model_paths should be supported by all custom nodes that load external models. I'm running in a docker environment.