自制LoRA模型分享以及使用教程

Author： Alpha
发布时间：April 5, 2023
2235 views
No comments
7171 words
Categories： python 编程技术番剧

训练素材大多数来自番剧截图以及中珂院图库再加上我自己的一点收藏。

素材质量可以说是没有达到那种训练教程的要求。

但是实际上对于仅仅训练特人物的Lora模型来说，这些训练素材已经足够了。

## 训练方式及脚本

[https://colab.research.google.com/drive/1_f0qJdM43BSssNJWtgjIlk9DkIzLPadx?usp=sharing&amp;pli=1#scrollTo=ZXFX2-C_Z-9N](https://colab.research.google.com/drive/1_f0qJdM43BSssNJWtgjIlk9DkIzLPadx?usp=sharing&pli=1#scrollTo=ZXFX2-C_Z-9N "https://colab.research.google.com/drive/1_f0qJdM43BSssNJWtgjIlk9DkIzLPadx?usp=sharing&amp;pli=1#scrollTo=ZXFX2-C_Z-9N")

没错，紧靠云端即可完成训练！

该训练方式需要您有可靠的魔法即可，把你的训练素材上传到谷歌网盘，之后按照提示启动，等待即可！

不过我的跑图是在本地完成的，包括我的打标签。

裁剪工具：[https://www.birme.net/?target_width=512&amp;target_height=512](https://www.birme.net/?target_width=512&target_height=512 "https://www.birme.net/?target_width=512&amp;target_height=512")

训练素材大小全部都是512*512（我习惯了，毕竟一开始embed就是这样训练的）

## 模型下载及使用

<button class=" btn m-b-xs btn-info btn-addon" onclick="window.open('https://blog.a152.top/usr/uploads/2023/04/753263622.safetensors','_blank')"><i class="glyphicon glyphicon-tags"></i>珂朵莉lora模型</button>

<button class=" btn m-b-xs btn-dark  " onclick="window.open('https://ply.lanzoub.com/io3280s4bcof','_blank')">一键下载三份lora（蓝奏云）</button>

### sdwebui的lora使用参见b站

sdwebui的搭建教程也从那个up主的投稿里面找吧！

[https://www.bilibili.com/video/BV1Py4y1d7eJ/?spm_id_from=333.337.search-card.all.click](https://www.bilibili.com/video/BV1Py4y1d7eJ/?spm_id_from=333.337.search-card.all.click "https://www.bilibili.com/video/BV1Py4y1d7eJ/?spm_id_from=333.337.search-card.all.click")

### 关于提示词

我提供的三份模型，提示词几乎是平常的那些词（只要你写的是女性相关的tag）最好还符合那么一点人物特征即可！可以参考我下面具体给出的例子！

最简单的触发词就是1gril,solo等等。

## 模型测试

也许有可能你会觉得换不了衣服什么的，实际上这是可以换的，你仅仅需要多写一些提示词即可！后面我也会给出一些实例

lora的模型权重如图所示

模型选用final-pure，也就是sd官方的模型（经过裁剪的）

### 奈芙莲lora测试 xy图

![请输入图片描述](http://img.shelterol.top/i/2023/ai/xy_grid-0082-1454611955-1girl,%20solo,%20twintails,%20ribbon,%20grey%20eyes.jpg)

提示词如下

```
1girl, solo, twintails, ribbon, grey eyes
```

```
(worst quality, bad quality:1.4), lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
```

Steps: 30, Sampler: Euler a, CFG scale: 9, Seed: 1454611955, Size: 512x512, Model hash: 89d59c3dde, Model: final-prune

### 艾瑟雅lora测试 xy图

![请输入图片描述](http://img.shelterol.top/i/2023/ai/xy_grid-0083-355953981-girl,%20solo,%20animal%20ears,%20braid,%20twin%20braids,%20_3,%20uniform.jpg)

```
girl, solo, animal ears, braid, twin braids, :3, uniform
```

Steps: 30, Sampler: Euler a, CFG scale: 9, Seed: 355953981, Size: 512x512, Model hash: 89d59c3dde, Model: final-prune

### 珂朵莉lora测试 xy图

![请输入图片描述](http://img.shelterol.top/i/2023/ai/xy_grid-0085-53592575-1girl,bangs,%20blue%20eyes,%20blue%20hair,%20eyebrows%20visible%20through%20hair,%20hair%20between%20eyes,%20long%20hair.jpg)

```
1girl,bangs, blue eyes, blue hair, eyebrows visible through hair, hair between eyes, long hair
```

```
 (worst quality, bad quality:1.4), lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
```

Steps: 30, Sampler: Euler a, CFG scale: 9, Seed: 53592575, Size: 512x786, Model hash: 89d59c3dde, Model: final-prune

下面的是512*512的，去掉了eyebrows这一条,增加了一些设定

![请输入图片描述](http://img.shelterol.top/i/2023/ai/xy_grid-0089-4042772542-1girl,bangs,%20blue%20eyes,%20blue%20hair,hair%20between%20eyes,%20long%20hair,open%20mouth,smile.jpg)

```
1girl,bangs, blue eyes, blue hair,hair between eyes, long hair,open mouth,smile
```

### 单张实例

**负面词都是一样的,见前面**

```
1girl, solo, twintails, ribbon, grey eyes,maid,peace symbol
```

Steps: 30, Sampler: Euler a, CFG scale: 9, Seed: 627951995, Size: 512x786, Model hash: 89d59c3dde, Model: final-prune,

![请输入图片描述](http://img.shelterol.top/i/2023/ai/00086-2382647111-1girl,%20solo,%20twintails,%20ribbon,%20grey%20eyes,maid,peace%20symbol.png)

</div><div role="tabpanel" id='tabs-e28c6b70d63b5dc14c5a3e78e3625e2d991' class="tab-pane fade  ">

```
1girl, solo, twintails, ribbon, grey eyes,revealing dress,on the beach
```

Steps: 30, Sampler: Euler a, CFG scale: 9, Seed: 928411891, Size: 512x768, Model hash: 89d59c3dde, Model: final-prune

![请输入图片描述](http://img.shelterol.top/i/2023/ai/00102-3185056920-1girl,%20solo,%20twintails,%20ribbon,%20grey%20eyes,revealing%20dress,on%20the%20beach.png)

</div><div role="tabpanel" id='tabs-c56a3dbbf368135f70aac1522480a1cd1002' class="tab-pane fade  ">

如果你发现你的动作，服装，环境很难控制时，多半是你的lora模型权重拉太高了，这是全标素材直接喂去训练成lora的通病。以及，出图多半有些运气成分在里面。

```
1girl, solo, twintails, ribbon, grey eyes,(whitedress),on the beach
```

![请输入图片描述](http://img.shelterol.top/i/2023/ai/xy_grid-0094-582317736-1girl,%20solo,%20twintails,%20ribbon,%20grey%20eyes,(whitedress),on%20the%20beach.png)

1girl, solo, twintails, ribbon, grey eyes,(whitedress),on the beach
Negative prompt: (worst quality, bad quality:1.4), lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
Steps: 30, Sampler: Euler a, CFG scale: 9, Seed: 582317736, Size: 512x768, Model hash: 89d59c3dde, Model: final-prune, ENSD: 31337, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: naifulianLORA-000003(deb386f74fae), AddNet Weight A 1: 0.65, AddNet Weight B 1: 0.65, X Type: AddNet Weight 1, X Values: "0.65,0.7,0.75,0.8"

</div>
</div>
</div>

让我们来看看其他的例子吧~~

### 珂朵莉

```
1girl,bangs, blue eyes, blue hair,wedding_dress,ptcrowd birds and cloud at sky
```

![请输入图片描述](http://img.shelterol.top/i/2023/ai/00145-1195925191-1girl,bangs,%20blue%20eyes,%20blue%20hair,wedding_dress,ptcrowd%20birds%20and%20cloud%20at%20sky.png)

![请输入图片描述](http://img.shelterol.top/i/2023/ai/00141-857834628-1girl,bangs,%20blue%20eyes,%20blue%20hair,wedding_dress,ptcrowd%20birds%20and%20cloud%20at%20sky.png)

### 艾瑟雅

```
1girl, solo, animal ears, braid, twin braids, :3,cat ears, long hair, brown eyes,orange hair,(wedding_dress)

```

![请输入图片描述](http://img.shelterol.top/i/2023/ai/00151-2333455275-1girl,%20solo,%20animal%20ears,%20braid,%20twin%20braids,%20_3,wedding_dress,ptcrowd%20birds%20and%20cloud%20at%20sky.png)

```
1girl, solo, animal ears, braid, twin braids, open mouse,smile,cat ears, long hair, brown eyes,((wedding_dress)),(forest),river,orange hair

```

![请输入图片描述](http://img.shelterol.top/i/2023/ai/00176-604470031-1girl,%20solo,%20animal%20ears,%20braid,%20twin%20braids,%20open%20mouse,smile,cat%20ears,%20long%20hair,%20brown%20eyes,((wedding_dress)),(forest),river,.png)

Last modification：April 5, 2023

如果觉得我的内容对你有用，请随意赞赏

自制LoRA模型分享以及使用教程

Alpha • 2023 年 04 月 05 日