This is a tweaker I originally trained as a TI embedding for 2.0, and 2.1. Every prompt I generated used it, but the data was lost, and this is as close as a LoRA can get to what that was. This LoRA will work with more models than the old TI could, and, as a matter of a fact, this works extremely well with other models more than base XL it was trained on.
All images shown are without and with done on the SDXL 1.0 base. Left side without and right side with (I only use the PC and images are shown two at a time).
Do NOT be afraid to play with the weight from -3 to about +5.