Not mine. https://civitai.com/models/189110?modelVersionId=212407
IMPORTANT: Version 2 is not necessarily a direct 'upgrade' from version 1. Rather it is an 'alternative' version, and you may obtain the best results by using both for different parts of a workflow.
LoRA Strengths Needed for Each Model
Version 1 requires a LoRA strength of 0.4.
Version 2 requires a LoRA strength of 0.1.
Things will go very wrong if you deviate more than half a point or so from these recommended strengths.
Usage
You may find that text-to-image generations using the more versatile V2 model can do more things, and it can do many of them better. When inpainting faces, it can sometimes produce a superior result to V1 - but remember that V1 was trained only on real data, while V2 has many synthetic faces that may not be totally 'realistic', and that this can sometimes be reflected in the output.
Therefore I recommend experimenting with both these models.
I also recommend using these LoRAs with the default SD V1.5 pruned checkpoint.
