I don't get it: the new models still need to be used somewhere, unless you're coding everything yourself, and that means they'll mostly get used in the updated ComfyUI nodes that work with the new checkpoints and LoRAs (and are available today).

Me either. Acly's krita add-on takes a bit of the pain out of it for me personally: https://github.com/Acly/krita-ai-diffusion (I have not and would not try FLUX however -I don't have the hardware).