Animagine XL 动漫角色模型

Animagine XL 4.0，也被风格化为Anim4gine，是终极动漫主题微调 SDXL 模型，也是Animagine XL 系列的最新版本。尽管是延续，但该模型是从Stable Diffusion XL 1.0重新训练而来的，使用来自各种来源的 8.4M 多样化动漫风格图像的海量数据集，知识截止时间为 2025 年 1 月 7 日，并经过大约 2650 GPU 小时的微调。与之前的版本类似，该模型使用标签排序方法进行身份和风格训练。

模型详细信息

• 开发者：Cagliostro 研究实验室
• 模型类型：基于扩散的文本到图像生成模型
• 许可证：CreativeML Open RAIL++-M
• 模型描述：这是一个可以根据文本提示生成并修改特定动漫主题图像的模型
• 微调自：Stable Diffusion XL 1.0

下游用途

1. 在我们的中使用这个模型Hugging Face Spaces
2. 使用它在ComfyUI或Stable Diffusion Webui
3. 与以下产品一起使用Cagliostro Colab Forge
4. 与🧨一起使用diffusers

🧨 扩散器安装

1.安装所需的库

pip install diffusers transformers accelerate safetensors --upgrade

2.示例代码

以下示例使用lpw_stable_diffusion_xl管道，可以更好地处理较长、加权和详细的提示。该模型已以 FP16 格式上传，因此无需variant="fp16"在from_pretrained调用中指定。

import torch
from diffusers import StableDiffusionXLPipeline

pipe = StableDiffusionXLPipeline.from_pretrained(
    "cagliostrolab/animagine-xl-4.0",
    torch_dtype=torch.float16,
    use_safetensors=True,
    custom_pipeline="lpw_stable_diffusion_xl",
    add_watermarker=False
)
pipe.to('cuda')

prompt = "1girl, arima kana, oshi no ko, hoshimachi suisei, hoshimachi suisei \(1st costume\), cosplay, looking at viewer, smile, outdoors, night, v, masterpiece, high score, great score, absurdres"
negative_prompt = "lowres, bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry"

image = pipe(
    prompt,
    negative_prompt=negative_prompt,
    width=832,
    height=1216,
    guidance_scale=6,
    num_inference_steps=25
).images[0]

image.save("./arima_kana.png")

使用指南

1. 提示结构

该模型使用基于标签的字幕和标签排序方法进行训练。使用以下结构化模板：

1girl/1boy/1other, character name, from which series, everything else in any order.

2. 质量增强标签

在提示的开始或结束处添加这些标签：

masterpiece, high score, great score, absurdres

3. 推荐负面提示

lowres, bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry

4. 最佳设置

• CFG 等级：5-7（建议 6）
• 采样步骤：25-28（建议25）
• 首选采样器：欧拉祖先（欧拉 a）

5. 建议的解决方案

方向 尺寸 长宽比

1024 x 1024 1:1

1152 x 896 9：7

1216 x 832 3：2

1344 x 768 7：4

1536 x 640 12：5

896 x 1152 7：9

832 x 1216 2：3

768 x 1344 4：7

640 x 1536 5:12

6. 最终提示结构示例

masterpiece, high score, great score, absurdres, 1girl, firefly \(honkai: star rail\), honkai \(series\), honkai: star rail, casual, solo, looking at viewer, outdoors, smile, reaching towards viewer, night

特殊标签

该模型支持各种特殊标签，可用于控制图像生成过程的不同方面。这些标签经过精心加权和测试，可在不同的提示中提供一致的结果。

品质标签

质量标签是直接影响整体图像质量和细节水平的基本控制。可用的质量标签：

• masterpiece
• best quality
• low quality
• worst quality

"masterpiece, best quality"使用质量标签并将负面提示留空的示例图像。使用"low quality, worst quality"质量标签并将负面提示留空的示例图像。

分数标签

与基本质量标签相比，分数标签可以对图像质量进行更细致的控制。它们对本模型中的转向输出质量有更大的影响。可用的分数标签：

• high score
• great score
• good score
• average score
• bad score
• low score

"high score, great score"使用分数标签并将负面提示留空的示例图像。使用"bad score, low score"分数标签并将负面提示留空的示例图像。

时间标签

时间标签允许您根据特定时间段或年份影响艺术风格。这对于生成具有特定时代艺术特征的图像非常有用。支持的年份标签：

• year 2005
• year {n}
• year 2025

带有时间标签的初音未来样本图像。 带有时间标签"year 2007"的初音未来样本图像。"year 2023"

作品参照

masterpiece, best quality, high score, great score, absurdres, 1girl, souryuu asuka langley, neon genesis evangelion, eyepatch, red plugsuit, sitting, on throne, crossed legs, head tilt, holding weapon, lance of longinus \(evangelion\), cowboy shot, depth of field, faux traditional media, painterly, impressionism, photo background
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry
Steps: 25, CFG scale: 6, Sampler: Euler a

masterpiece, best quality, high score, great score, absurdres, pointy ears, 1girl, frieren, sousou no frieren, long hair, 1boy, flower, head wreath, smile, petals, earrings, jewelry, shirt, elf, blue flower, looking at another, outdoors, green eyes, upper body, long sleeves, striped shirt, striped, sky, solo focus, closed mouth, white hair, falling petals
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry
Steps: 25, CFG scale: 6, Sampler: Euler a

masterpiece, best quality, high score, great score, absurdres, 1girl, mita \(miside\), miside, blue eyes, low twintails, red hairband, blue scrunchie, red hairclip, red choker, finger to mouth, partially shaded face, head tilt, gesugao, smirk, upper body, dark background, dark
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry
Steps: 25, CFG scale: 6, Sampler: Euler a, Seed: 1173448321

1girl/1boy, character name, from what series, everything else in any order.

Quality Modifier	Score Criterion
masterpiece	        > 95%
best quality	        > 85% & ≤ 95%
great quality	        > 75% & ≤ 85%
good quality	        > 50% & ≤ 75%
normal quality	        > 25% & ≤ 50%
low quality	        > 10% & ≤ 25%
worst quality	        ≤ 10%

Rating Modifier	    Rating Criterion
safe	            General
sensitive	    Sensitive
nsfw	            Questionable
explicit, nsfw	    Explicit

Year Tag	Year Range
newest	        2021 to 2024
recent	        2018 to 2020
mid	        2015 to 2017
early	        2011 to 2014
oldest	        2005 to 2010

Aesthetic Tag	       Score Range
very aesthetic	       > 0.71
aesthetic	       > 0.45 & < 0.71
displeasing	       > 0.27 & < 0.45
very displeasing       ≤ 0.27

nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]

masterpiece, best quality, very aesthetic, absurdres

Dimensions	Aspect Ratio
1024 x 1024	1:1 Square
1152 x 896	9:7
896 x 1152	7:9
1216 x 832	19:13
832 x 1216	13:19
1344 x 768	7:4 Horizontal
768 x 1344	4:7 Vertical
1536 x 640	12:5 Horizontal
640 x 1536	5:12 Vertical

1girl, souryuu asuka langley, neon genesis evangelion, rebuild of evangelion, lance of longinus, cat hat, plugsuit, pilot suit, red bodysuit, sitting, crossed legs, black eye patch, throne, looking down, from bottom, looking at viewer, outdoors, masterpiece, best quality, very aesthetic, absurdres
Negative prompt: wings, nsfw, low quality, worst quality, normal quality,
Steps: 28, Size: 896x1152, Seed: 3651045455, Model: animagine-xl-3.1, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: Euler a, CFG scale: 7, Model hash: 9f86289b6a, Hires steps: 15, Hires upscale: 1.5, Hires upscaler: Latent (nearest-exact), Denoising strength: 0.55