A negative LoRA is meant to be used with a strength less than 0. The idea is that you are getting the “opposite” of what it was trained on.

You can find my SDXL Negative LoRA here.
And my Stable Diffusion 2.1 768 Embedding here.

Here is a Stable Diffusion XL render with out the LoRA:

This is the same settings with the LoRA added at -1.0 strength:

As you can see, more details and a more finished result.

Here is -2.0:

This is an easy way to keep the same general composition of the seed while adding more complex details to the image.