This LoRA was trained on cherry-picked images generated with OhMen Toontastic with my previous LoRA, which was taken down. It did not work very well because it was trained on a small set of close-up photos. This new version should work much better in realistic scenarios where the generation is more about the subject than just the necklace.