Notice
V3 has better performance than V2, and V2 has significantly better performance than V1.
Compatibility
Trained on the NAI model, and is therefore compatible with any anime-styled model such as Anything, Counterfeit, AOM, and PastelMix. This is shown in the example images.
Note that the cross-shaped pupils show up best in Anything and AOM.
Features
As of V3, the LoRA is able to do the following.
Dress up Nahida in alternate outfits. Examples of how to do this are in the sample images
Responds very well to perspective prompts such as from below, from behind, etc.
Give Nahida her signature cross-shaped pupils. This occurs whenyou generate at a sufficiently large resolution(above 896x896). Examples are given in the sample images
Can inpaint in the perfect looking cross-shaped pupils
Inpainting
If you have generated a good Nahida image and wish to add the pupils in via inpainting, simply mask the face and inpaint with the following settings
prompt: masterpiece, best quality, 1girl, solo, nahida_genshin, cross-shaped pupils <lora:NahidaV3:1>
neg-prompt: EasyNegative, extra fingers, fewer fingers
Default inpainting settings, 25 steps, 832x832 resolution, and 0.55 denoising strength. Make sure to select inpaint area as "Only masked"
This should add her cross-shaped pupils in a few tries.
Recommended Weight
In V3, generating at weight = 1.0 produces good results for all occasions.
模型为转载，如有侵权，请联系删除
