Quoted By:
Really not seeing the appeal of SDXL.
It's slow, wastes data on vague knowledge of characters and styles that is much better left to LoRa, speaking of barely has any LoRas.
And everyone's outputs look the same.
There are some niche things it does a bit better in terms of posing, probably due to the broader dataset.
If switching between XL and 1.5 wasn't such a pain in the ass it might be worth using XL to gen base images and then fixing them in 1.5 (or even the inverse for certain small details like eyes, which it seems quite good at), but at least on Vlad switching requires a full restart and changing several settings.