Buy me a can of cat food - buymeacoffee.com/Darkseal

This textual inversion was not made from images of Luna Lovegood or Evanna Lynch, rather it was trained from images created from a mix of ckpts merged with the Luna data found here https://civarchive.com/models/1161/luna

Evanna Lynch's features seem to want to push stable diffusion to make her face rounder, giving her younger characteristics. Because of that I didn't feel right trying to release any kind of checkpoint merge.

I'm not quite sure which step version is "the best", have not gotten a chance to play around with it too much yet. All of these samples has Restore Face OFF. 9 times out of 10 it will not help. But, if you wanted someone that looks "close", but not Luna, then turn it on. It was very difficult picking out "one" to use. All the versions past 200 steps looked fine to me. 3000 might be a smidge over baked, idk. Have fun! Keep it classy. Check out more examples of this and other models here https://www.deviantart.com/darksealstudios