玩转 Stable Diffusion WebUI 各类模型

Stable Diffusion WebUI 最有意思的地方不是在安装好之后生成图像,而是各种各样的模型。

提前警告:如果你的硬盘空间不够大的话,还是不要随便玩模型了,随随便便就是好几 G,又得甚至于 10 多个 G。

目前我仅留了最常用的 SD V1.5 和 SD V2.1 两个模型,大小为 13G。

notion image

另外还需要说明一点,就是我曾经测试过用 NAS 来存储模型使用,完全不能用,暂时没有时间具体去研究到底什么原因。只有老老实实的继续在本地硬盘上跑。所以 NAS 上存了大量模型,真需要用到的时候再复制过来。

写这篇文章也是因为近期玩模型过程中打算整理一波,一是方便自己,二么也算是对其他小伙伴做些贡献。

Stable Diffusion 各种模型层出不穷,要说完估计需要费一番功夫,所以我摒弃其他小模型,只整理收集大模型,就是 ckpt 和 safetensors。如果你也打算跟着我一起玩模型但是还未安装,可以先参看我之前的文章:

在 Apple Silicon M1/M2 Mac 上安装和运行 Stable Diffusion

说实话,我找了好多关于如何在 M1/M2 上安装和运行 Stable Diffusion 的教程和帖子,发现相互之间借鉴的不少,但是能用的确实没几个。 寻找一番后,发现其实没那么复杂。也不知道为什么网上的那么多教程搞得那么复杂,又是这个又是那个的一大堆,简单实现的方式有好几种:

https://www.hivan.me/How%20to%20install%20and%20run%20Stable%20Diffusion%20on%20Apple%20Silicon

在 Apple Silicon M1/M2 Mac 上安装和运行 Stable Diffusion

还是先从最基础的模型开始:

Stable Diffusion

其他多数模型基本上都是从这个基础模型上再次训练得到的。

Stable Diffusion v2.1

  1. SDv2.1 提升了人物生成能力,因为 SDv2.0 大量增加了风景、建筑物和动物的数据集,减少了人物的学习量。

  2. SDv2.1 提高了 NSFW 过滤器准确度,因为 SDv2.0 的成人过滤器过滤的太狠,错误判定很多

  3. 即使是极端长宽比的图像也能顺利生成。

  4. 解剖学的身体和手(特别是手掌)的描写精度提高。

512 X 512 model :

stabilityai/stable-diffusion-2-1-base · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/stabilityai/stable-diffusion-2-1-base

stabilityai/stable-diffusion-2-1-base · Hugging Face

768 X 768 model:

stabilityai/stable-diffusion-2-1 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/stabilityai/stable-diffusion-2-1

stabilityai/stable-diffusion-2-1 · Hugging Face

img2img model

stabilityai/stable-diffusion-2-depth · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/stabilityai/stable-diffusion-2-depth

stabilityai/stable-diffusion-2-depth · Hugging Face

重绘 model

stabilityai/stable-diffusion-2-inpainting · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/stabilityai/stable-diffusion-2-inpainting

stabilityai/stable-diffusion-2-inpainting · Hugging Face

超分 model

stabilityai/stable-diffusion-x4-upscaler · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/stabilityai/stable-diffusion-x4-upscaler

stabilityai/stable-diffusion-x4-upscaler · Hugging Face

Stable Diffusion V 1.5

runwayml/stable-diffusion-v1-5 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main

runwayml/stable-diffusion-v1-5 at main

Stable Diffusion V 1.4

CompVis/stable-diffusion-v-1-4-original · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/CompVis/stable-diffusion-v-1-4-original

CompVis/stable-diffusion-v-1-4-original · Hugging Face

NovelAI

大名鼎鼎的 NovelAI,属于商业泄露模型。经过人在回路精细微调,可以生成高质量的二次元图像。但是千万时刻记得这个可是商用泄露模型,要注意避免法律风险:

pub-2fdef7a2969f43289c42ac5ae3412fd4.r2.dev

https://pub-2fdef7a2969f43289c42ac5ae3412fd4.r2.dev/animefull-latest.tar

Waifu Diffusion

基于 Stable Diffusion 模型训练得到,增加了动漫及人物训练得到的模型,基本平时各种公开场合看到 WD 就是他。

WD 和 NovelAI 模型有些同质化,但是 NovelAI 实际是商用模型泄露,在某些使用情况下是有风险的。而 WD 不是,不过也不是说他绝对安全,毕竟 WD 也使用 Danbooru 进行学习,所以如果你关心这个需要注意一点。

Waifu Diffusion V1.5

这个模型使用是需要一个 yaml 文件的,究其原因是这个模型是基于 SD V2 得出的,需要把和 Model 同名的 yaml 文件放在模型所在的文件夹下,目前 1.5 模型是 beta2 版本,持续迭代 ing…

Waifu Diffusion v1.5 beta

waifu-diffusion/wd-1-5-beta · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/waifu-diffusion/wd-1-5-beta

waifu-diffusion/wd-1-5-beta · Hugging Face

VAE(1.4 VAE 通用)

vae/kl-f8-anime2.ckpt · hakurei/waifu-diffusion-v1-4 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt

vae/kl-f8-anime2.ckpt · hakurei/waifu-diffusion-v1-4 at main

YAML

waifu-diffusion/wd-1-5-beta2 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/waifu-diffusion/wd-1-5-beta2/tree/main/checkpoints

waifu-diffusion/wd-1-5-beta2 at main

Waifu Diffusion V1.4

和 1.5 版本一样,基于 SD V2 得到的,依然需要下载 yaml 文件放在 model 同文件夹下。

Waifu Diffusion V 1.4

wd-1-4-anime_e1.ckpt · hakurei/waifu-diffusion-v1-4 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/wd-1-4-anime_e1.ckpt

wd-1-4-anime_e1.ckpt · hakurei/waifu-diffusion-v1-4 at main

wd-1-4-anime_e2.ckpt · hakurei/waifu-diffusion-v1-4 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/wd-1-4-anime_e2.ckpt

wd-1-4-anime_e2.ckpt · hakurei/waifu-diffusion-v1-4 at main

VAE(1.5 通用)

vae/kl-f8-anime2.ckpt · hakurei/waifu-diffusion-v1-4 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt

vae/kl-f8-anime2.ckpt · hakurei/waifu-diffusion-v1-4 at main

YAML

e2 和 e1 是通用的,但是需要改名

hakurei/waifu-diffusion-v1-4 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hakurei/waifu-diffusion-v1-4/tree/main

hakurei/waifu-diffusion-v1-4 at main

*Elysium Anime*

生成偏真实风格的动漫图片,风格比较偏向西式,光影还不错。

模型推荐写下面这些负面提示,可有效提升质量。

1
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

Elysium_V1

偏真实风的模型,手画的还不错,模型底稿基本是以西方人为主,所以生成的脸也偏西方人。

Elysium_V1.ckpt · hesw23168/SD-Elysium-Model at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hesw23168/SD-Elysium-Model/blob/main/Elysium_V1.ckpt

Elysium_V1.ckpt · hesw23168/SD-Elysium-Model at main

*SD_Elysium_Kuro_Model*

与 Anything 4.0、WD 1.4 等合并后经过微调的二次元用模型。已经包含 WD 的“kl-f8-anime2”VAE 文件,因此无需使用额外的 VAE 文件

Elysium_Kuro_Anime_V1.safetensors · hesw23168/SD_Elysium_Kuro_Model at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hesw23168/SD_Elysium_Kuro_Model/blob/main/Elysium_Kuro_Anime_V1.safetensors

Elysium_Kuro_Anime_V1.safetensors · hesw23168/SD_Elysium_Kuro_Model at main

*Elysium_Anime_V3*

动漫的附加学习模型,NSFW 化相当严重,有更清晰的轮廓和轻微的三维效果。基于 Elysium_V1

Elysium_Anime_V3.safetensors · hesw23168/SD-Elysium-Model at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/hesw23168/SD-Elysium-Model/blob/main/Elysium_Anime_V3.safetensors

Elysium_Anime_V3.safetensors · hesw23168/SD-Elysium-Model at main

*Anything 系列*

Anything 是个神奇的二次元模型,据说是基于几十种模型融合+未知图片训练而来,随便写几个提示,就能到的不错的结果。不过这个模型整个就是一团混沌,实际训练模型,过程,方法,作者全部都是未知的。模型容易过拟合,非专业人士,请不要在此基础上训练模型。

Anything v3.0

“应该”是基于 NAI 模型+WD+SD 等几十种模型+手部图片强化训练得出的。实际训练模型,过程,方法,作者全部都是未知的。如果没有.vae.pt,图片整体颜色浓度(饱和度)会更很浅。PS:Anything v3.0 的 .vae.pt 文件可以用于 NAI。

  • Anything V3.0 fp16: magnet:?xt=urn:btih/:45cd353ac4fa87098db5e3a6a349539710a3a1f5&dn=Anything-V3.0-fp16.zip

  • Anything v3.0 fp32: magnet:?xt=urn:btih/:d9db662ab5ace77004b3348c23c9381380c27156&dn=Anything-V3.0-fp32.zip

  • Anything v3.0 full-ema: magnet:?xt=urn:btih/:80460036625fb61dce4bc6e7dab744744309a2a0&dn=Anything-V3.0-fullema.zip

huggingface.co

https://huggingface.co/Linaqruf/anything-v3-better-vae/tree/main

Anything v4

自称是 Anything 最新版本的模型,实际一切都是未知的。仅需几个提示即可生成详细的 2D 插图的能力以及使用 danbooru 标签的能力。整体比过拟合的 v3 更自然,人物姿势等更容易操作。

anything-v4.0-pruned.safetensors · andite/anything-v4.0 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/andite/anything-v4.0/blob/main/anything-v4.0-pruned.safetensors

anything-v4.0-pruned.safetensors · andite/anything-v4.0 at main

Anything v4.5

貌似是 Anything v4 的进化,但实际一切都是未知的。比 v4 画风更柔和一点。

anything-v4.5-pruned.safetensors · andite/anything-v4.0 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/andite/anything-v4.0/blob/main/anything-v4.5-pruned.safetensors

anything-v4.5-pruned.safetensors · andite/anything-v4.0 at main

Zeipher

生成更符合真人解剖结构的真人模型,训练集以女性图像为主官方网站是 https://ai.zeipher.com,已经关闭。请不要用真人模型画明星和未成年的 NSFW 内容,不然你可能会遇到很麻烦的法律问题

F222

f222.safetensors · acheong08/f222 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/acheong08/f222/blob/main/f222.safetensors

f222.safetensors · acheong08/f222 at main

F111

f111.ckpt · Reachout/F111 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/Reachout/F111/blob/main/f111.ckpt

f111.ckpt · Reachout/F111 at main

3DKX

因为 Zeipher 官方已经 GG,这是热心网友创建的衍生 3DKX 模型如果你想让你的 3D 角色有一张更“二次元”的脸,提示词最开始写 “3d cartoon of”,或者如果你想要经典的 3D 渲染外观,写“a 3d render of”高分辨率模型,推荐分辨率为 1152 x 768 或更高

3DKX_1.0b

f111.ckpt · Reachout/F111 at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/Reachout/F111/blob/main/f111.ckpt

f111.ckpt · Reachout/F111 at main

R34

从网站“rule34.xxx”的 150,000 张图像中进行训练。rule34.xxx 几乎全是 NSFW 图片,所以你懂的

r34_e4

1.99 GB file on MEGA

1.99 GB file on MEGA

https://mega.nz/file/yJgDUCzA#zOD2yeE6QLBqPEjEpIi2b4FWOlb64yVUveOd_eW6teI

1.99 GB file on MEGA

磁力链接:magnet:?xt=urn:btih/:ed9f0e3f849d7119107ef4e072c6abeb129e1a51&dn=r34_e4.ckpt

EVT pixiv 排行榜模型

基于 pixiv 排行图片训练,夹杂有部分 R18 排行图片

Evt_V4_e10_ema

Evt_V4_e10_ema.safetensors · haor/Evt_V4-preview at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/haor/Evt_V4-preview/blob/main/Evt_V4_e10_ema.safetensors

Evt_V4_e10_ema.safetensors · haor/Evt_V4-preview at main

EVT_V3

haor/Evt_V3 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/haor/Evt_V3

haor/Evt_V3 · Hugging Face

EVT_V2

haor/Evt_V2 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/haor/Evt_V2

haor/Evt_V2 · Hugging Face

Basil_mix

逼真的真人模型,基于亚洲风格训练,支持 Danbur 标签提示词需要加载 VAE,不然画面色彩浓度和边缘会很淡提示词应尽可能简单不要堆砌大量质量标签和负面提示,不然会适得其反。请不要用真人模型画明星和未成年的 NSFW 内容,不然你可能会遇到很麻烦的法律问题

basil_mix

basil mix.ckpt · nuigurumi/basil_mix at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/nuigurumi/basil_mix/blob/main/basil%20mix.ckpt

basil mix.ckpt · nuigurumi/basil_mix at main

VAE

vae-ft-mse-840000-ema-pruned.ckpt · stabilityai/sd-vae-ft-mse-original at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt

vae-ft-mse-840000-ema-pruned.ckpt · stabilityai/sd-vae-ft-mse-original at main

Chillout Mix

逼真的真人模型,基于亚洲风格训练,支持 Danbur 标签提示词请不要用真人模型画明星和未成年的 NSFW 内容,不然你可能会遇到很麻烦的法律问题

chillout mix _ NiPruned Fp32 Fix

chilloutmix_NiPrunedFp32Fix.safetensors · Inzamam567/useless_Chillout_mix at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

https://huggingface.co/Inzamam567/useless_Chillout_mix/blob/main/chilloutmix_NiPrunedFp32Fix.safetensors

Uber Realistic Porn Merge

如名字所说,逼真的真人 Porn 模型,简称 URPM 模型请不要用真人模型画明星和未成年的 NSFW 内容,不然你可能会遇到很麻烦的法律问题

Uber Realistic Porn Merge

Uber Realistic Porn Merge (URPM) | Stable Diffusion Checkpoint | Civitai

For early access builds and to support daily work on URPM, please check out my patreon! https://www.patreon.com/uber_realistic_porn_merge , or disc...

Uber Realistic Porn Merge (URPM) | Stable Diffusion Checkpoint | Civitai

https://civitai.com/models/2661/uber-realistic-porn-merge-urpm

玩转 Stable Diffusion WebUI 各类模型

https://hivan.me/stable-diffusion-webUI-models/

作者

Hivan Du

发布于

2023-03-09

更新于

2024-01-16

许可协议

评论