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 <a target="_blank" rel="ugc" href="https://twitter.com/KovarShark">artist</a>.

Kovar Style - SeaArt

Trained with 'kovar' in the dataset, I think it helps SeaArt 'align' with the artist style. More saturated than v1 (which is accurate), but more consistent style.

Files with SHA256: 2e2647dcef42555a384a8d071dcba8823549f82765dc868851b9420cc6512148