v1 is not trained with 'kovar' as a tag in the dataset, but v2 is.

Model weight of 0.6-1.0 works, but some tags might feel de-emphasized at 1.0, so 0.8 is fine.

And of course don't forget to support the artist.