[English / 中文]

Suggestion: Elaina is a complicated character who has many details. You’d better follow the instruction in order to avoid chaotic details. It is highly recommended to view all example images before you start using.

Basic Information

Base model: Animefull-latest

Recommended weight: 0.7 - 0.9

You could use these trigger words to generate particular elements of Elaina (of course you can add a few modifiers such as color):

Hair: long / short hair, floating hair, braid, hair bundle

Hair ornaments: yellow hair band, hair bow, pink bow

Witch hat: hat, witch hat, black headwear

Shirt: shirt, white shirt, collared shirt

Necktie: yellow bowtie, neck bowtie, ribbon, neck ribbon

Witch clothes: witch, robe, black robe, open robe, wide sleeves, black skirt

Other clothes: dress, capelet, jacket, pajamas etc.

Boots: boots, black footwear, pointy footwear

Other ornaments: badge, wand, broom, animal ears, thighhighs

Expressions: shaded face, closed / open eyes, closed / open mouth, sweat drop, smug, happy, angry, sad, nervous, surprised, scared, smile, :d

Advanced Skills

Those are basic trigger words. The example images show more details. I’m sorry for my insufficient competence to make this LoRA perfect. You may witness much chaos when you are using the model, especially when incorporating elements such as badge, wand, broom etc. You could try to add some tags above into negative prompts or adjust LoRA weight to solve the problem. Some prompts tend to be used at the same time at training datasets, therefore using these prompts will cause problems.

In this LoRA I would stick to the idea that a successful character LoRA should reflect his / her personalities. For this reason I did not devote all my efforts into Elaina's clothes and ornaments. As you can see, there have been LoRAs in civitai that do well in this.

As a matter of fact, there are many different painting styles of Elaina that reflect her different personalities. I trained two LoRA whose datasets are from cover images of the novel and the animation respectively. I uploaded 3 files including these two versions (in zip pack). The final version is the mixed one. You could use them properly to produce better results. For example, the cover version can be used to fix some detail problems, such as the wrong magic badge (brooch). Also, you can use the cover version to upscale images generated by the mixed version so that the results will perform well both in overall layout and details.

Example Image Introduction

The example images (originally 600 * 900 definition, 2x SD upscale) are based on Anything V5 PrtRE and AnimeVAE. Due to the complexity of the character, some example images have to be manually refined (PS + inpainting). Some materials have been uploaded in example images for your manual improvement.

1-5. witch clothes

6-9. other clothes and hair styles

11-13. different poses and angles

13-17. expressions

18. comparison among different versions and among different checkpoints

19. some materials (brooch, broom and wand)

Acknowledgements

It is my second LoRA and I've learned more to improve its quality compared to the first one. Nevertheless, the complexity causes many problems. It will be highly appreciated if you could stand with them. Thanks for your downloads and support!

建议：鉴于伊蕾娜拥有许多复杂的装饰细节，为了避免细节混乱，请参照教程使用本模型，并在使用之前最好先参考一遍所有例图。

基本信息

训练底模: Animefull-latest

推荐权重: 0.7 - 0.9

建议使用以下触发词来生成伊蕾娜的特定元素（当然你可以给这些触发词再添加一些像颜色这样的修饰）:

头发： long / short hair, floating hair, braid, hair bundle

发饰： yellow hair band, hair bow, pink bow

魔女帽 hat, witch hat, black headwear

衬衫 shirt, white shirt, collared shirt

领带： yellow bowtie, neck bowtie, ribbon, neck ribbon

魔女服：witch, robe, black robe, open robe, wide sleeves, black skirt

其它服饰： dress, capelet, jacket, pajamas 等

靴子： boots, black footwear, pointy footwear

其它装饰： badge, wand, broom, animal ears, thighhighs

神态： shaded face, closed / open eyes, closed / open mouth, sweat drop, smug, happy, angry, sad, nervous, surprised, scared, smile, :d

进阶技巧

以上是基本的触发词，例图展示了更多使用细节。我很抱歉我的能力有限，不能制作一个完美的LoRA，所以你在使用中可能会遇到不少图片的细节混乱，尤其是使用诸如胸针、魔杖、扫帚等元素的时候。你可以尝试将上面的内容加入负面提示词或者调整LoRA权重来解决问题。一些提示词在训练集中一般是成对出现的，因此单独使用它们可能会带来问题。

在这个LoRA中我仍然坚持LoRA应当反应人物气质的理念，因此我并没有花过多精力去还原伊蕾娜的服饰。何况，在civitai中已经有许多优秀的LoRA做到了这一点。

实际上，伊蕾娜有非常多的画风，每种画风都有自己的特色。我分别选择了小说封面以及动漫作为素材训练了2种LoRA，最终版是这两个版本的融合版本，而独立版本在上传的zip文件中。你可以灵活使用它们以达到更好的效果，比如你可以用封面版本的LoRA去修复像胸针这样的细节。或者你可以先用融合或动漫版本生成图片，再用封面版本进行图生图放大，这样可以兼顾整体构造与细节。

例图介绍

例图原始分辨率为600*900，用了2倍的SD放大，底模Anything V5 PrtRE，使用AnimeVAE。由于人物复杂性，一些展示的例图经过了一定的人工细节修正（PS+重绘）。在例图中有一些素材文件可供你进行人工修正。

1-5. 魔女服

6-9. 各类服饰及发型

10-13. 各类姿势及角度

14-17. 各类表情

18. 不同数据集版本、不同checkpoint大模型的比较

19. 一些素材（胸针、扫帚和魔杖）

致谢

这是我做的第二个LoRA，相比第一个模型又有一定进步，但是伊蕾娜是一个更加复杂的角色，请多多包涵不足之处。感谢您的下载与支持！