I calculated what combination of SDXL models can produce a general purpose model that can produce near perfect results while being one model. I've already made something like this for SD1.5, but I never published it.

RATIOS-CLiP:

24.75% SDXL_NIJIse

20.25% DynaVisionXL

55% CopaxTimelessXL_V7

RATIOS-UNET:

50% CopaxTimelessXL_V7

25% JuggernautV2

12.5% SDXL_NIJIse

12.5% DynaVisionXL