Stefan picked SDXL for its ecosystem, not because it is the newest model.
LoRAs and ControlNets 'need to be actually developed for that main model'. An add-on built for one base model will not work on another. If you use SDXL, you need SDXL LoRAs and SDXL ControlNets. Before relying on a ControlNet type such as depth or canny, check that an SDXL version exists (depth-and-canny-controlnet-maps-from-input).
When the goal is structural control, as in Blender to ComfyUI SDXL image-to-image render pipeline, the available add-ons matter more than raw base-model quality. This choice is time-sensitive. Re-check add-on coverage for newer models before switching. See Juggernaut XL checkpoint and the image-to-image sampler graph.