Seeking Advice on Persistent Model Loading in ComfyUI

[mzhangzhicheng]

I'm currently facing challenges with the repeated loading times of models in ComfyUI. Each time I need to use a model, it requires a complete reload, which is quite time-consuming.

Is there a way to configure ComfyUI to maintain the models in a loaded state, perhaps through certain parameters or settings? This would greatly enhance efficiency and usability.

Any guidance or tips on this would be highly appreciated.

Thank you!

[tiago-falves]

Hey! Did you find a solution?

[mzhangzhicheng]

No, I just reloaded models.

[ZodiacFRA]

I have the same problem here, using SDXL turbo and the ComfyUI API. The most time consuming part is model reloads at every request, even if the prompt json uses the same model everytime:

Requested to load AutoencoderKL Loading 1 new model Prompt executed in 3.54 seconds Requested to load SDXL Loading 1 new model

Any solutions / workarounds?