<pre><code># Ogipote style lora

Base Model: Anima 1.0 Base

## 示例图

画廊所有图片均可以直接导入 ComfyUI

文字 `CHENING` 是模型直出的，并没有经过 PS 修改

## 建议参数

一些经验参数从 NetaLumina Yume 过渡来的，还是有点用的

### 正面提示词

随便，如果你不需要 Pony 油，则不建议打 `score_*` 之类的玩意

Anima 对纯 Tag 支持还是好很多的，对英文NL遵循还行，对中文NL支持很差

### 负面提示词

固定用这个，LLM 系 text encoder 天然对自然语言输入有良好支持，包括负面

这是本人总结的一点经验，可看你们自己

```
worst quality, low quality, score_1, score_2, score_3, artist name,

This image suffers from low resolution and poor quality, appearing pixelated, blurry, and full of compression or jpeg artifacts. 

The scene is cluttered with many small figures, heads, or people in the background that look like chibi characters, dolls, or fumo dolls, surrounding the subject with a large collection of plush toys. 

The artwork looks like a fake anime or game screenshot with noise, errors, and scan lines, resembling something unfinished sketch with a simple, abstract, or monochrome background.

The anatomy is incorrect, featuring bad proportions, bad perspective, and ugly, deformed body parts such as bad hands, bad feet, fused or missing fingers, extra limbs, and twisted or flexible deformities. The image contains four fingers, six fingers, four toes, six toes.

This image contains many of the artist's signatures around the edges, and contains many watermarks, signatures, logos, or censorship bars like mosaics, and has an low-score AI-generated error.

The character in this image has a white area above the eyes, a white forehead, and many lines on the eyelids.

Draw by @Alexander Groznov, @Andrew Wyeth, @Boris Vallejo, @Entei Ryu, @Manara Milo, @Marco Turini, @Norman Rockwell, @Paulina Sadowska, @Syd Mead, @Syd Mills, @THORES Shibamoto, @chinese anime style, @renwan.
```

### 采样

Step：30
CFG：4.5 ~ 5.5
Sampler：seeds_2 / seeds_3 (容易煎炸) / er_sde / ...
Scheduler：simple / normal / ...

### Latent 放大

Width * Height：2048 * 2048 (保持宽高比)
Upscale method：lanczos
Step：8
CFG：3.0
Sampler：res_multistep
Schedule：simple
Denoise：0.18 (重要)

### 画师串

随便，一般推荐 `多LoRa + 串` 混合模式出图，远比 `单LoRa` 或 `单串` 效果要好

### LoRa 堆叠

随便，一般推荐 `多LoRa + 串` 混合模式出图，远比 `单LoRa` 或 `单串` 效果要好

可以堆叠一些稳固手的 LoRa

## 其他技术支持和探索

NetaYume4 方面：https://civarchive.com/models/2014460/netalumina-xcn-light-style

## 限制

### 用户承诺

- ​禁止用于违法活动​（如欺诈、诽谤、侵犯隐私、传播仇恨言论等）；

- ​遵守所在地法律法规，包括但不限于数据安全、版权法及AI伦理规范；

- ​不得绕过模型内置安全机制​（如内容过滤规则）。

### 作者明确不承担以下责任​

- 因用户使用生成内容导致的直接或间接损失​（如法律纠纷、经济损失）；

- 生成内容的准确性、适用性及道德合规性；

- 因模型漏洞、数据偏差导致的输出错误或偏见​。
</code></pre><p></p>

Files with SHA256: 8b1d6f5674e4a883c904b22cc59b152a7e5d608c27947797e28dba4e4c9b01fe