触发词(Trigger word): @afezeria
底模(base model): MiaoMiao Harem(1.2) (https://civarchive.com/models/934764/miaomiao-harem?modelVersionId=3020110)
使用NoobAl(Anishadow_v2)+afezeria画风Lora生成图片，筛选出我想要的Afezeria的某个时期的画风，扩充原始数据集，并对原始数据集进行清洗，删除其中部分画风不匹配的图片，训练分辨率从1024上升到1280
部分数据集参考：https://www.pixiv.net/artworks/146411910
推荐采样器：er_sde或euler，其中er_sde采样器更符合我想要的画风。例图中第二张和第七张使用euler采样器，其余图片均使用er_sde采样器。
由于我对数据集进行了筛选并添加了我想要的画风的图片，因此该Lora风格可能与原始Afezeria画风有所出入。
虽然效果有所提升，但我看来还是达不到在NoobAI上的效果(参考https://www.pixiv.net/artworks/146411910)。
This LoRA was trained using NoobAI (Anishadow_v2) together with the original Afezeria style LoRA. I selected images that matched a specific period of Afezeria's style that I wanted, expanded the original dataset, and cleaned it by removing images whose style did not match the target style. The training resolution was increased from 1024 to 1280.
Part of the dataset was referenced from: https://www.pixiv.net/artworks/146411910
Recommended samplers: er_sde or euler. Among them, er_sde produces a style that is closer to what I am looking for. In the example images, the second and the seventh image were generated using the euler sampler, while all other images were generated using er_sde.
Since I filtered the dataset and added images that matched my preferred version of the style, the resulting LoRA may differ somewhat from Afezeria's original style.
Although the results have improved, I still feel that it does not achieve the same quality as NoobAI's rendering of the style (reference: https://www.pixiv.net/artworks/146411910).
如果你喜欢这个Lora，请不妨点个赞或者使用它发布作品
If you like my Lora. please give it a thumbs up or use it to publish your work.