Warning: Those don't work with Civitai On-site generator. Once again, i don't understand why some embeddings are ok and some others are not. It feels like it due to their size: if they go over a limit of number of tokens, they are just ignored 😥

I still haven't done the LoRA for my OC, but nonetheless, here are the embeddings, i have used until now to generate her. They work nicely, the only drawbacks is that they are a LOT of tokens internally: a LoRA would allow having a prompt more focused on everything else in the picture composition.

• LiadoraDescILL (39 tokens): The base description of Liadora (hair and eyes colors, the tattoo and horns)

• LiadoraOutfitILL (54 tokens): Her standard outfit

• LiadoraNegILL (26 tokens): The "fix" negative to avoid some influence from 2B (white hair and Yorha outfit)

(I have added the number of tokens corresponding to each embeddings. Remember that for any SDXL models and derivative, a prompt is split and ingested 75 by 75 tokens).

A base prompt:

Positive: 1girl, LiadoraDescILL, LiadoraOutfitILL

Negative: IllusN3g,  LiadoraNegILL

The result (bonus trick: if needed, you can increase the weight of the embedding, like LiadoraNegILL:1.1)

A more complex prompt may need to split the embeddings into two (at least) chunks manually using BREAK to avoid side effects (and that's why i'll need to make a LoRA out of those 😅):

1girl, solo, LiadoraDescILL, sitting on a throne, indoor, castle, (conjuring a blazing purple fireball with one hand)
BREAK
LiadoraOutfitILL, from below
BREAK
volumetric lighting, black background, simple background, vibrant, stunningly beautiful, crisp, detailed, sleek, ultramodern, suble shadows, high contrast, cinematic, ultra detailed, intricate, professional