Reloading model on each request, doubling generate time

[mtourne]

I'm running a relatively simple API workflow with IPAdapter and a hi-res fix using sdxl-turbo

When I'm running the same workflow inside of ComfyUI each generation takes around .7 seconds on my setup.
When I run it via TDComfyUI it takes around 2 seconds

And the SDXL model is reloaded after each prompt is received :

got prompt
Requested to load SDXL
Loading 1 new model
100%|████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  9.33it/s]
100%|████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  8.63it/s]
Prompt executed in 2.44 seconds

Here is the api workflow I use :
dat_ipdadapter_hires__td_5064_19.json

I'm running it on windows 11 with an rtx 4090

PS. in comparison I've tried a simpler img2img workflow with sdxl-turbo and the reloading doesn't take place :
dat_sdxl_turbo_img2img_ext_api__td_5064_5.json