Ear down helper

"Ear down" tag helper will help improve the capabilities of anime models to make characters with one animal ear pointed downward. It has minimal style bleed as can be seen in the XY plot comparing different models. A bit over half of images generated will have the desired one ear down, sometimes neither or both ears will be down.

Strength: 0.6-1
Trigger words: ear down
Other keywords: animal ears, cat ears, fox ears, wolf ears, bunny ears, horse ears, deer ears
Trained on model: NovelAI

The preview images are made with the AnyLoRA checkpoint since it is a good neutral anime model to showcase LoRAs with. The training data consists of 111 "close-up" 512x512 images.

Usage tips: in order to not get a face only image you will need to use other tags to change the composition, such as "full body", "upper body", "standing", etc. See the preview images at the end for examples.