diff --git a/config_ivonblog.toml b/config_ivonblog.toml index bdbd918..111c98a 100644 --- a/config_ivonblog.toml +++ b/config_ivonblog.toml @@ -51,20 +51,19 @@ enableGitInfo = true [menu] [[menu.before]] - name = "【回Ivon的部落格】Back to Ivon's Blog" + name = "【回Ivon的部落格】" url = "https://ivonblog.com" weight = 10 [[menu.before]] - name = "【意見回饋】Send feedbacks" - url = "https://github.com/ivon852/stable-diffusion-webui-manuals" + name = "【看更多Stable Diffusion教學】" + url = "https://ivonblog.com/categories/stable-diffusion教學/" weight = 20 [[menu.after]] name = "© 2023 Ivon Huang. 文章授權:創用CC-姓名標示-相同方式分享" url = "https://github.com/ivon852/stable-diffusion-webui-manuals" - weight = 10 - + weight = 20 [params] # (Optional, default light) Sets color theme: light, dark or auto. diff --git a/content.zh-cn/_index.md b/content.zh-cn/_index.md index ff5cbd8..b3ec42e 100644 --- a/content.zh-cn/_index.md +++ b/content.zh-cn/_index.md @@ -4,7 +4,7 @@ bookFlatSection: true title: "欢迎" --- -这本非官方的中文使用手册是针对AUTOMATIC1111开发的Stable Diffusion WebUI所撰写,对应版本为**v1.6.0** +这本非官方的中文使用手册是针对AUTOMATIC1111开发的Stable Diffusion WebUI所撰写,对应版本为**v1.6.0**。最后更新日期如网页底部所示。 内容有:介绍如何将Stable Diffusion WebUI安装至您的电脑,以及讲解相关AI绘图功能用法,最后再介绍模型训练方式。 diff --git a/content.zh-cn/extensions/_index.md b/content.zh-cn/extensions/_index.md index 3a15d3d..fea5cc5 100644 --- a/content.zh-cn/extensions/_index.md +++ b/content.zh-cn/extensions/_index.md @@ -6,6 +6,8 @@ bookCollapseSection: true Stable Diffusion WebUI除了基本的文生图、图生图、内补绘制之外,还能安装扩充功能(Extensions,或称「插件」),在绘图的时候集成其他AI技术,或是简化操作,生成更好的作品。 -这一章介绍设计给Stable Diffusion WebUI的扩充功能用法,并附上一些范例。包含界面中文化、控制动作的ControlNet、生成多个人物的Latent Couple等等。 +例如:界面中文化、控制动作的ControlNet、生成多个人物的Latent Couple的扩充功能等等。 + +因为扩充功能实在太多,这一章仅介绍Stable Diffusion WebUI的扩充功能装法,各个扩充功能的细节部份不多谈。 针对Stable Diffusion WebUI设计的扩充功能非常多,本章列表不定时更新。 diff --git a/content.zh-cn/extensions/a1111-sd-webui-tagcomplete.md b/content.zh-cn/extensions/a1111-sd-webui-tagcomplete.md deleted file mode 100644 index a910631..0000000 --- a/content.zh-cn/extensions/a1111-sd-webui-tagcomplete.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "提示词自动补完 Booru tag autocompletion" -weight: 5 ---- - -- 扩充功能来源:[DominikDoom/a1111-sd-webui-tagcomplete](https://github.com/DominikDoom/a1111-sd-webui-tagcomplete) - -输入提示词的时候提示相关Danbooru上的标签,画动漫图的时候很有用,还可以安装中文标签翻译。 - -![](../../../images/a1111-sd-webui-tagcomplete-1.webp) - -进入Extensions页面,点击`Install from URL`,输入网址`https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git`,按`Install`,之后重启WebUI。 - -如何安装标签中文翻译:到[byzod/a1111-sd-webui-tagcomplete-CN](https://github.com/byzod/a1111-sd-webui-tagcomplete-CN)保存库,点击绿色的Code → Download ZIP,解压缩,得到tags文件夹,下面有`Tags-zh-full-pack.csv`和`config.json`这二个文件。将这二个文件移动到`stable-diffusion-webui/extensions/a1111-sd-webui-tagcomplete/tags`文件夹。 diff --git a/content.zh-cn/extensions/auto-sd-paint-ext.md b/content.zh-cn/extensions/auto-sd-paint-ext.md deleted file mode 100644 index c0feb72..0000000 --- a/content.zh-cn/extensions/auto-sd-paint-ext.md +++ /dev/null @@ -1,56 +0,0 @@ ---- -title: "绘图软件Krita集成 auto-sd-paint-ext" -weight: 10 ---- - -- 扩充功能来源:[Interpause/auto-sd-paint-ext](https://github.com/Interpause/auto-sd-paint-ext) - -这是将Stable Diffusion的Inpainti集成至绘图软件Krita的扩充功能,可在选取图层范围后调用Stable Diffusion填充内容,支持不规则选取。 - -1. 进入Extension页面,点击Install from URL:`https://github.com/Interpause/auto-sd-paint-ext.git`,关闭SD WebUI。 - -2. 编辑SD WebUI的启动参数,加入`--api`启动SD WebUI。进入网页界面会看到多了`auto-sd-paint-ext Guide/Panel`页面 - -![](../../../images/auto-sd-paint-ext-1.webp) - -3. 安装自由软件[Krita](https://krita.org/) - -4. 启动Krita,点击上方的设置 → 管理资源,点击「打开资源文件夹」 - -![](../../../images/auto-sd-paint-ext-2.webp) - -5. 拷贝`pykrita`文件夹的路径 - -![](../../../images/auto-sd-paint-ext-3.webp) - -6. 回到SD WebUI,粘贴pykrita文件夹的路径,下面即会产生创建符号链接的指令。打开终端机,按照你的操作系统运行指令。 - -![](../../../images/auto-sd-paint-ext-4.webp) - -7. 重新启动Krita。点击设置 → 设置Krita,点击左边的「Python插件程序管理器」,再勾选右边的Stable Diffusion Plugin - -![](../../../images/auto-sd-paint-ext-5.webp) - -8. 重新启动Krita,编辑图片时左边即会多出一堆皮肤。这些皮肤可从设置 → 工具皮肤叫出来 - -![](../../../images/auto-sd-paint-ext-6.webp) - -9. 以Inpaint的功能为例子,让我们先把无关的皮肤都关掉,只留下"Inpaint" - -![](../../../images/auto-sd-paint-ext-7.webp) - -10. 点击右下角添加颜料图层 - -![](../../../images/auto-sd-paint-ext-8.webp) - -11. 在此图层用铅笔工具,将要填充的地方涂黑 - -![](../../../images/auto-sd-paint-ext-9.webp) - -12. 于左边的皮肤填入提示词,点击Start Inpaint,开始绘图。进度条可从终端机查看。 - -![](../../../images/auto-sd-paint-ext-10.webp) - -13. SD WebUI运算完成后结果即会显示在Krita的图层上。 - -![](../../../images/auto-sd-paint-ext-11.webp) diff --git a/content.zh-cn/extensions/controlnet.md b/content.zh-cn/extensions/controlnet.md deleted file mode 100644 index de786b0..0000000 --- a/content.zh-cn/extensions/controlnet.md +++ /dev/null @@ -1,223 +0,0 @@ ---- -title: "ControlNet 按照骨架动作绘图" -weight: 3 ---- - -- 扩充功能来源:[Mikubill/sd-webui-controlnet](https://github.com/Mikubill/sd-webui-controlnet) -- ControlNet的保存库:[lllyasviel/ControlNet](https://github.com/lllyasviel/ControlNet) - -目前版本:ControlNet v1.1 - -ControlNet是通过加入额外条件来控制扩散模型的神经网络结构,它可以让AI参考给定图片的动作/线条/景深,更精准的生成图片。 - -跟内置的「图生图」技术比起来,ControlNet的效果更好,能让AI以指定动作生图;再搭配3D建模作为辅助,能缓解单纯用文生图手脚、脸部表情画不好的问题。 - -ControlNet的用法还有:上传人体骨架线条,ControlNet就能按骨架的动作生成完稿的人物图片。或是上传素色的3D建模,让ControlNet彩现成为室内布置家具。 - -Lvmin Zhang是ControlNet原始程序的开发者,Mikubill则是开发扩充功能,让我们可以在Stable Diffusion WebUI用ControlNet生图。 - - -# 1. 安装ControlNet与下载模型 - -1. 切换至Extensions页面,点击`Install From URL`,URL输入`https://github.com/Mikubill/sd-webui-controlnet.git`,按下Install,接着从终端机按CTRL+C,关闭Stable Diffusion WebUI。 - -2. 到[lllyasviel/ControlNet v1.1](https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main)下载以`.pth`结尾的模型档,全部文件加起来约18GB。`.yaml`档不需要下载。目前v1.1是测试版,之后会合并回[主保存库](https://huggingface.co/lllyasviel/ControlNet)。 - -![](../../../images/controlnet-1.webp) - -3. 将这些模型档放到`stable-diffusion-webui/extensions/sd-webui-controlnet/models`文件夹 - -4. 重新启动Stable Diffusion WebUI。 - - -# 2. 各个模型的用途 - -ControlNet现有以下模型,您可以按照需求只下载需要的模型。 - -## Anime Lineart - -侦测线条,生成的图片亦会保留原始的线条,适合处理动漫图像 - -## Canny - -侦测图片边缘,比较模糊,不如Scribbles完整。 - -## Depth - -侦测输入图片的深度图(depth map)。 - -## Illumination - -侦测输入图片的光源与照明效果。 - -## Inpaint - -功能类似「内补绘制」,使用50%随机遮罩+50%随机光流遮罩训练而成。 - -## Instruct Pix2Pix - -模型文件名为`ip2p`,类似「图生图」,但是使用训练50%的指示(instruction)提示词和50%的叙述(description)提示词训练而成。因为是ControlNet,使用此模型时不需要调整CFG Scale。 - -根据原作者的说法,此模型在下「使其成为X」的提示词所生成的图,效果比「使Y成为X」要好。 - -> Also, it seems that instructions like "make it into X" works better than "make Y into X". - -## Lineart - -侦测线条,适合处理线稿,生成的图片亦会保留原始的线条。 - -## M-LSD - -侦测输入图片的直线。 - -## Normalbae - -## Openpose - -使用OpenPose技术侦测输入图片人物的动作,不一定会保留线条。 - -## Scribbles - -侦测线条,侦测到的线条品质介于Soft Edge和Lineart之间。 - -## Segmentation - -模型文件名为`seg`,将侦测的图片对象切成一个一个色块处理,例如房子一个色块,后面的天空一个色块。 - -## Shuffle - -把输入图片的概念转移到生成的图片。 - -作者给的例子:输入灰色装甲图片,生成的钢铁人盔甲也会是灰色的。 - -## Soft Edge - -侦测图片边缘,效果较为柔和,像用炭笔涂过。 - -## Tile - -输入图片,选取一个区域,使其变清晰的模型。 - - -# 3. 参数解说 - -在安装之后,WebUI的界面会多出"ControlNet"的按钮。 - -![](../../../images/controlnet-2.webp) - -点开会看到以下选项: - -## Single Image - -上传单张图片。 - -## Batch - -处理多张图片。Input Directory填写的是待处理的图片所在路径。 - -## Open New Canvas - -清除图片,开一张新画布。 - -## 使用电脑镜头 - -在Open New Canvas的选项下面有四个按钮。 - -![](../../../images/controlnet-3.webp) - -由左到右:添加画布;使用电脑镜头拍照上传;未知;还原上次算图设置。 - -## Enable - -在生图时激活ControlNet,必选。 - -## Low VRAM - -生图使用ControlNet会吃更多VRAM,此选项可降低VRAM使用量。 - -## Pixel Perfect - -由ControlNet自动决定Preprocessor分辨率。 - -## Allow Preview - -允许预览生图结果,会显示该模型侦测的线条或骨架。 - -## Preprocessor 预处理器 - -通常Preprocessor和Model二者要一致。 - -Preprocessor是先将上传的图片处理过一轮,例如`Scribbles`会将彩色图片转成线稿。若图片的线条是白底黑线,且不是用WebUI画布所绘制,则Preprocessor必须选取"Invert"。 - -如果你的图片不需要处理,Preprocessor可留空。 - -## Run Preprocessor (爆炸图标) - -运行预处理器,预先处理图片。 - - -## Model 模型 - -设置生图的时候使用哪个ControlNet模型。 - -## Control Weight - -ControlNet在生图时的权重。 - -## Starting Control Step - -要让ControlNet开始参与生图的步数。 - -## Ending Control Step - -让ControlNet结束参与生图的步数。 - -## Preprocessor Resolution - -预处理器的分辨率。 - -如果不知道要设多少,请勾选"Pixel Perfect"让ControlNet自动为您决定。 - -## Control Mode (Guess Mode) - -控制生图时,要AI平衡二者(Balanced),或是偏重你的提示词(My prompt is more important),还是偏重ControlNet (ControlNet is more important)。 - -## [Loopback] Automatically send generated images to this ControlNet unit - -自动将生成的图片传回此ControlNet运算单元(unit)。 - -是的,如果您的电脑VRAM够大,您可以在Settings → ControlNet激活多个ControlNet运算单元。 - -## Multi ControlNet - -一个ControlNet效果不够,您有没有想过激活第二、第三个ControlNet!? - -缺省安装后只会使用一个ControlNet,但是叠加多个有助于生成更好的图片。比方说同时使用Open Pose和Depth生成人体,并防止手画歪。 - -要激活Multi ControlNet,请进入Settings → ControlNet,拖动拉杆,设置要激活的ControlNet数量。 - - - - -# 4. 实际操作 - -此处以改绘一个人物姿势为例子。 - -1. 在安装之后,WebUI的界面会多出"ControlNet"的按钮。 - -![](../../../images/controlnet-5.webp) - -2. 进入文生图的页面填入提示词,接着点开下面ControlNet的界面,勾选`Enabled`激活ControlNet,上传图片。勾选LowVRAM可降低VRAM占用。 - -![](../../../images/controlnet-6.webp) - -3. 接着选取要使用的Preprocessor和Model,二者要一致。 - -Preprocessor是先将上传的图片处理过一轮,例如`Scribbles`会将彩色图片转成线稿。如果你的图片不需要处理,则Preprocessor可留空。 - -![](../../../images/controlnet-7.webp) - -4. 以Scribbles为例,选取后点击右上角Generate即会生成出结果,并附上侦测到的线条。如下图所示,右边贞德的姿势确实跟左边的02十分类似。 - -![](../../../images/controlnet-8.webp) - diff --git a/content.zh-cn/extensions/easynegative.md b/content.zh-cn/extensions/easynegative.md deleted file mode 100644 index 80f8c57..0000000 --- a/content.zh-cn/extensions/easynegative.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "简易负向提示词 EasyNegative" -weight: 4 ---- - -- 扩充功能来源:[gsdf/EasyNegative](https://huggingface.co/datasets/gsdf/EasyNegative) - -用这个取代冗长的负向提示词吧,反过来利用Embedding的特性,告诉AI不要画什么。 - -到[作者的保存库](https://huggingface.co/datasets/gsdf/EasyNegative)下载`EasyNegative.safetensors`,将它放到`stable-diffuison-webui`文件夹下的`embeddings`,重启WebUI。 - -之后在WebUI的负向提示词字段填入`EasyNegative`即会生效,可有效替代冗长的负向提示词。不过单独使用效果可能还是不好,你还是需要额外准备一些负向提示词。 - -![](../../../images/easynegative-1.webp) diff --git a/content.zh-cn/extensions/latent-couple.md b/content.zh-cn/extensions/latent-couple.md deleted file mode 100644 index 8cbd87f..0000000 --- a/content.zh-cn/extensions/latent-couple.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "生成多个人物 Latent Couple" -weight: 8 ---- - -- 扩充功能来源:[ashen-sensored/stable-diffusion-webui-two-shot](https://github.com/ashen-sensored/stable-diffusion-webui-two-shot) - -Latent Couple会分割绘图时的提示词,这样就可以生成多重人物/对象,并精确指定位置颜色了。 - -像是先用PoseX拉好骨架,再使用Latent Couple标出人物的概略位置。 - -于Extensions页面的Install from URL输入以下网址安装: `https://github.com/ashen-sensored/stable-diffusion-webui-two-shot.git`,再点击Reload WebUI。 - -安装后于文生图的界面即会看到Latent Couple的选项。 diff --git a/content.zh-cn/extensions/multidiffusion-upscaler-for-automatic1111.md b/content.zh-cn/extensions/multidiffusion-upscaler-for-automatic1111.md deleted file mode 100644 index b918897..0000000 --- a/content.zh-cn/extensions/multidiffusion-upscaler-for-automatic1111.md +++ /dev/null @@ -1,16 +0,0 @@ ---- -title: "低VRAM也能生大图 Tiled Diffusion" -weight: 5 ---- - -- 扩充功能来源:[pkuliyi2015/multidiffusion-upscaler-for-automatic1111](https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111) - -这个扩充功能让VRAM偏低(< 6GB)的电脑也能生成1080x1080以上分辨率的大型图片,甚至不需要用`--lowvram`和`--medvram`参数来启动Stable Diffusion WebUI了。 - -安装方法:进入Extensions页面,点击Install from URL,输入`https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111.git`后按Install,重启WebUI。 - -在文生图的页面左下角会看到Tiled Diffusion和Tiled VAE,将二者的Enabled勾选即为激活,生图时维持默认值即可。 - -![](../../../images/multidiffusion-upscaler-for-automatic1111-1.webp) - -另提供Region Prompt Control,可以控制每个tile的提示词。 diff --git a/content.zh-cn/extensions/openpose-for-blender.md b/content.zh-cn/extensions/openpose-for-blender.md deleted file mode 100644 index fc6e9de..0000000 --- a/content.zh-cn/extensions/openpose-for-blender.md +++ /dev/null @@ -1,12 +0,0 @@ ---- -title: "3D人体+手部+脸部模型 Openpose for blender" -weight: 12 ---- - -- 扩充功能来源:[Character bones that look like Openpose for blender _ Ver_8.1 Depth+Canny+Landmark+MediaPipeFace+finger](https://toyxyz.gumroad.com/l/ciojz) - -设计给Blender使用的3D人体模型,集成OpenPose和手部、脸部的模型,在Blender内摆好姿势后输出图片,即可搭配ControlNet生图。 - -*需要安装Blender 3.0以上版本。 - -作者开放自由捐款,不想花钱的话金额填0即可下载。 diff --git a/content.zh-cn/extensions/posex.md b/content.zh-cn/extensions/posex.md deleted file mode 100644 index 5a81a56..0000000 --- a/content.zh-cn/extensions/posex.md +++ /dev/null @@ -1,32 +0,0 @@ ---- -title: "骨架人偶 PoseX" -weight: 6 ---- - -- 扩充功能来源:[hnmr293/posex](https://github.com/hnmr293/posex) - -*需要先安装ControlNet才能使用这个扩充功能。 - -PoseX是可以在Stable Diffuison WebUI直接拉人物骨架,再配合ControlNet生成姿势的扩充功能。 - -类似的扩充功能为[OpenPose Editor](https://github.com/fkunn1326/openpose-editor),差别在于PoseX可以旋转放大。 - -1. 切换至Extensions页面,点击Install From URL,URL输入`https://github.com/hnmr293/posex.git`,按下Install。接着重启WebUI。 - -![](../../../images/posex-1.webp) - -2. 打开文生图的页面,点击右下角PoseX,点击Send this image to ControlNet - -![](../../../images/posex-2.webp) - -3. 在下面的ControlNet,点击Enabled,preprocessor选取`none`,model选`openpose`,不需要上传图片。 - -![](../../../images/posex-3.webp) - -4. 回到上面的PoseX,调整人物姿势。左键点击移动,滚轮放大缩小,对模型左键点二下即可用右键移动单个骨架。 - -![](../../../images/posex-4.webp) - -5. 填入正负向提示词,即会按照PoseX的姿势来生成图片。 - -![](../../../images/posex-5.webp) diff --git a/content.zh-cn/extensions/sd-extension-system-info.md b/content.zh-cn/extensions/sd-extension-system-info.md deleted file mode 100644 index 6f995af..0000000 --- a/content.zh-cn/extensions/sd-extension-system-info.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "即时监控系统信息 System Info" -weight: 5 ---- - -- 扩充功能来源:[vladmandic/sd-extension-system-info](https://github.com/vladmandic/sd-extension-system-info) - -此扩充功能给Stable Diffusion WebUI添加一个监控系统信息的页面,显示主程序版本、系统信息、GPU、RAM、使用的模型、使用的命令行参数等等。 - -![](../../../images/sd-extension-system-info-1.webp) - -点击下面的Run benchmark可以跑分测试系统性能,有助于评估主程序更新后性能是否有变化。点击Submit results将结果回报到[Vladimir Mandic的网站](https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html)。 - -![](../../../images/sd-extension-system-info-2.webp) diff --git a/content.zh-cn/extensions/sd-webui-depth-lib.md b/content.zh-cn/extensions/sd-webui-depth-lib.md deleted file mode 100644 index c41f34a..0000000 --- a/content.zh-cn/extensions/sd-webui-depth-lib.md +++ /dev/null @@ -1,34 +0,0 @@ ---- -title: "手部模型 Depth map library and poser" -weight: 7 ---- - -- 扩充功能来源:[jexom/sd-webui-depth-lib](https://github.com/jexom/sd-webui-depth-lib) - -此扩充功能加入手部模型,可配合PoseX的骨架使用,改善AI手部画不好的问题。 - -于Extensions页面的Install from URL输入以下网址安装: `https://github.com/jexom/sd-webui-depth-lib`,重启WebUI。 - -这里示范搭配PoseX画手的用法,请记得激活Multi ControlNet。 - -1. 于文生图页面点开PoseX,设置`Target ControlNet number`为1,摆好骨架,并设置。然后点击Save PNG,得到一张png图档 - -![](../../../images/sd-webui-depth-lib-1.webp) - -2. 切换至Depth map library页面,点击Add Background Image,上传PoseX骨架的图档。 - -![](../../../images/sd-webui-depth-lib-2.webp) - -3. 再点击左边各个手的模型,加到右边的图片,将二只手摆出来。再点Send to Img2img。 - -![](../../../images/sd-webui-depth-lib-3.webp) - -4. 回到文生图页面,点开ControlNet,将0号ControlNet的preprocessor和model设置为Depth,Weights设置为0.5。 - -![](../../../images/sd-webui-depth-lib-4.webp) - -5. 再点击1号ControlNet,将model设置为OpenPose,Weights设置为0.5。 - -![](../../../images/sd-webui-depth-lib-5.webp) - -6. 点击Generate生成,可看到图片按照方才制作的手和骨架生成了。 diff --git a/content.zh-cn/extensions/sd-webui-mov2mov.md b/content.zh-cn/extensions/sd-webui-mov2mov.md deleted file mode 100644 index 615014d..0000000 --- a/content.zh-cn/extensions/sd-webui-mov2mov.md +++ /dev/null @@ -1,27 +0,0 @@ ---- -title: "AI绘图转视频 mov2mov" -weight: 7 ---- - -- 扩充功能来源:[Scholar01/sd-webui-mov2mov](https://github.com/Scholar01/sd-webui-mov2mov) - -*需要先安装ControlNet才能使用这个扩充功能。 - -将视频逐一抽出画格,使用ControlNet生图,然后再自动合成新视频。可以设置输出的画格率,将人物单独处理。 - - -1. 切换到mov2mov页面,输入提示词, 再上传视频。 - -![](../../../images/mov2mov-1.webp) - -2. 在下面设置输出视频的长宽。生图的长宽比例应与原始视频一致。 - -![](../../../images/mov2mov-2.webp) - -3. 最下面是单独用ControlNet处理人物的选项,可以改善动作侦测的精度。 - -![](../../../images/mov2mov-3.webp) - -4. 点击右上角Generate,即会开始生成。成品位于主程序下的`/outputs/mov2mov-images`目录。 - -![](../../../images/mov2mov-4.webp) diff --git a/content.zh-cn/extensions/stable-diffusion-webui-civitai-helper.md b/content.zh-cn/extensions/stable-diffusion-webui-civitai-helper.md deleted file mode 100644 index a07aa08..0000000 --- a/content.zh-cn/extensions/stable-diffusion-webui-civitai-helper.md +++ /dev/null @@ -1,54 +0,0 @@ ---- -title: "自动从CivitAI下载模型和缩略图 Civitai Extension" -weight: 5 ---- - -- 扩充功能来源:[butaixianran/Stable-Diffusion-Webui-Civitai-Helper](https://github.com/butaixianran/Stable-Diffusion-Webui-Civitai-Helper) - -喜欢从[CivitAI](https://civitai.com/)下载模型的用户,装这个扩充功能就不用[手动下载模型](../installation/download-models)了!此扩充功能于Stable Diffusion WebUI的界面加入CivitAI模型下载页面,方便下载各种模型,还会自动比对数据抓取模型的缩略图。 - -安装方法:进入Extensions页面,点击Install from URL,输入`https://github.com/civitai/sd_civitai_extension.git`后按Install,关掉终端机,重启WebUI。 - -重开WebUI页面即会多出"Civitai Helper"。 - -![](../../../images/stable-diffusion-webui-civitai-helper-1.webp) - - -此页面有4个功能: - - -## 1. 自动下载缩略图 - -点击橘色的Scan按钮,它就会自动比对本机模型(存盘点、嵌入、超网络、LoRA)的数据,从CivitAI下载模型缩略图。 - -![](../../../images/stable-diffusion-webui-civitai-helper-2.webp) - - -## 2. 从CivitAI取得模型信息 - -此功能用于爬取本机模型的信息。 - -于`Civitai URL`的字段填入模型的网址(就是你用浏览器所看到的网址),例如这里我选择Deliberate这个本机模型。 - -点击`Get Model Info from Civitai`即会下载模型的数据。 - -![](../../../images/stable-diffusion-webui-civitai-helper-3.webp) - - -## 3. 从CivitAI下载模型档 - -1. 于`Civitai URL`字段填入模型的网址(就是你用浏览器所看到的网址),点击`1. Get Model Info by Civitai Url`取得该页面的数据。 - -![](../../../images/stable-diffusion-webui-civitai-helper-4.webp) - -2. 它会自动抓取模型种类和信息,`Model Name`模型名称和`Model Version`模型版本可以自选。点击`3. Download Model`开始下载模型。至于下载进度条请看终端机。 - -![](../../../images/stable-diffusion-webui-civitai-helper-5.webp) - - -## 4. 检查模型有无更新 - -将Model Types全部打勾,点击`Check New Version from Civitai`,若有新版模型它就会自动下载。 - -![](../../../images/stable-diffusion-webui-civitai-helper-6.webp) - diff --git a/content.zh-cn/extensions/stable-diffusion-webui-images-browser.md b/content.zh-cn/extensions/stable-diffusion-webui-images-browser.md deleted file mode 100644 index b3158c8..0000000 --- a/content.zh-cn/extensions/stable-diffusion-webui-images-browser.md +++ /dev/null @@ -1,18 +0,0 @@ ---- -title: "图片库浏览器 Images Browser" -weight: 5 ---- - -- 扩充功能来源:[yfszzx/stable-diffusion-webui-images-browser](https://github.com/yfszzx/stable-diffusion-webui-images-browser) - -安装后即可方便浏览生图的结果,并查看使用过的提示词历史纪录。 - -安装方法:进入Extensions页面,点击Install from URL,输入`https://github.com/yfszzx/stable-diffusion-webui-images-browser.git`后按Install,重启WebUI。WebUI页面即会多出"Images Browser"的页面。 - -点进去,按一下First Page即会显示最近生图的结果。 - -![](../../../images/stable-diffusion-webui-images-browser-1.webp) - -点击单张图片,查看使用过的提示词,还可以给予星级评分。 - -![](../../../images/stable-diffusion-webui-images-browser-2.webp) diff --git a/content.zh-cn/extensions/stable-gimpfusion.md b/content.zh-cn/extensions/stable-gimpfusion.md deleted file mode 100644 index f39b3b1..0000000 --- a/content.zh-cn/extensions/stable-gimpfusion.md +++ /dev/null @@ -1,30 +0,0 @@ ---- -title: "修图软件GIMP集成 Stable-Gimpfusion" -weight: 9 ---- - -- 扩充功能来源:[ArtBIT/stable-gimpfusion](https://github.com/ArtBIT/stable-gimpfusion) - -ArtBIT开发的Stable-Gimpfusion能让你在GIMP修图时的图层调用Stable Diffusion填充内容,支持文生图、图生图、ControlNet。 - -类似的扩充功能有[blueturtleai/gimp-stable-diffusion](https://github.com/blueturtleai/gimp-stable-diffusion),支持不规则选取区,还可以调用志愿者经营的stable-horde服务器来绘图。但是它不支持SD WebUI。 - -1. 编辑SD WebUI的启动参数,加入`--api`参数,再启动SD WebUI。网页界面可以不用打开。 - -2. 安装自由软件[GIMP](https://www.gimp.org/) - -3. 下载[stable_gimpfusion.py](https://raw.githubusercontent.com/ArtBIT/stable-gimpfusion/main/stable_gimpfusion.py),放到GIMP的扩充功能文件夹 (Linux为`$HOME/.gimp-2.10/plug-ins/`并赋予运行权限,Windows为`C:\Users\{your_id}\AppData\Roaming\GIMP\2.10\plug-ins\`) - -4. 打开GIMP,打开图片。 - -5. 点击右上角的Gimpfusion,点击要使用文生图(text to image)还是图生图(image to image) - -![](../../../images/stable-gimpfusion-1.webp) - -6. 输入提示词,Stable Diffusion就会开始生图 - -![](../../../images/stable-gimpfusion-2.webp) - -7. 进度条可从终端机查看,之后结果会添加至GIMP的图层。目前此扩充功能只支持填充整个图层。 - -![](../../../images/stable-gimpfusion-3.webp) diff --git a/content.zh-cn/extensions/text-generation-webui.md b/content.zh-cn/extensions/text-generation-webui.md deleted file mode 100644 index 4691ecb..0000000 --- a/content.zh-cn/extensions/text-generation-webui.md +++ /dev/null @@ -1,8 +0,0 @@ ---- -title: "大型语言模型集成 Text Generation WebUI" -weight: 11 ---- - -与AI对话,并利用Stable Diffusion生图。 - -参考 [Text Generation WebUI](https://ivonblog.com/posts/text-generation-webui-extensions/)的「sd_api_pictures」扩充功能。 diff --git a/content.zh-cn/installation/command-line-arguments-and-settings.md b/content.zh-cn/installation/command-line-arguments-and-settings.md index 7a48271..cfdf83b 100644 --- a/content.zh-cn/installation/command-line-arguments-and-settings.md +++ b/content.zh-cn/installation/command-line-arguments-and-settings.md @@ -3,9 +3,7 @@ title: "2.3. 命令行参数" weight: 2 --- -{{< hint danger >}} -此页面仅供参考:左边各个系统(Linux/Windows/macOS)的安装教学已包含懒人参数,照抄即可。 -{{< /hint >}} +注:这篇文章仅供参考,本系列教学已经附上预先设置好的指令参数了,照抄即可。 Stable Diffusion WebUI的项目文件夹附有启动主程序的指令稿(script),称为启动指令稿。 @@ -30,6 +28,8 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram 可在启动指令稿使用这些环境变量(environment variables): +
+ |名称|说明| |---|---| |PYTHON|设置Python运行档路径。| @@ -45,6 +45,7 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram |SD_WEBUI_CACHE_FILE|缓存文件路径,默认值为根目录下的`cache.json`。| |SD_WEBUI_RESTAR|由启动指令稿(`webui.sh`或`webui.bat`)设置的值,告诉WebUI可以使用重启功能。| |SD_WEBUI_RESTARTING|检测WebUI是否正在重启或重新加载的内部数值,可以用来关闭自动打开浏览器的功能。设为`1`即不要自动打开浏览器。设为`0`的话,即使WebUI正在重启也自动打开浏览器。| +
# 3. 命令行参数 @@ -53,65 +54,66 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram 注意下面是二条横线「- -」 +
-| 参数指令 | 数值 | 默认值 | 说明 | +| 参数指令|数值|默认值|说明 | |---|---|---|---| | **设置值** | -|-h, --help | None | False | 显示此帮助消息并退出。 | -|--exit | | | 安装后终止程序。 | -|--data-dir | DATA_DIR | ./ | 用户数据保存的路径。 | -|--config | CONFIG | configs/stable-diffusion/v1-inference.yaml | 建构模型设置档的路径。 | -|--ckpt | CKPT | model.ckpt | Stable Diffusion模型的存盘点模型路径。一旦指定,该模型会加入至存盘点模型列表并加载。| -|--ckpt-dir | CKPT_DIR | None | 存放Stable Diffusion模型存盘点模型的路径。 | -|--no-download-sd-model | None | False | 即使找不到模型,也不自动下载SD1.5模型。 | -|--vae-dir | VAE_PATH | None | VAE的路径。 | -|--gfpgan-dir| GFPGAN_DIR | GFPGAN/ | GFPGAN路径 | -|--gfpgan-model| GFPGAN_MODEL | GFPGAN模型文件名 | -|--codeformer-models-path | CODEFORMER_MODELS_PATH | models/Codeformer/ | Codeformer模型档的路径。 | -|--gfpgan-models-path | GFPGAN_MODELS_PATH | models/GFPGAN | GFPGAN模型档的路径。 | -|--esrgan-models-path | ESRGAN_MODELS_PATH | models/ESRGAN |ESRGAN模型档的路径。| -|--bsrgan-models-path | BSRGAN_MODELS_PATH | models/BSRGAN |BSRGAN模型档的路径。 | -|--realesrgan-models-path | REALESRGAN_MODELS_PATH | models/RealESRGAN | RealESRGAN模型档的路径。 | -|--scunet-models-path | SCUNET_MODELS_PATH | models/ScuNET | ScuNET模型档的路径。 | -|--swinir-models-path | SWINIR_MODELS_PATH | models/SwinIR | SwinIR和SwinIR v2模型档的路径。 | -|--ldsr-models-path | LDSR_MODELS_PATH | models/LDSR | 含有LDSR模型档的路径。 | -|--lora-dir | LORA_DIR | models/Lora | 含有LoRA模型档的路径。| -|--clip-models-path | CLIP_MODELS_PATH | None | 含有CLIP模型档的路径。 | -|--embeddings-dir | EMBEDDINGS_DIR | embeddings/ | Textual inversion的embeddings路径 (缺省: embeddings) | -|--textual-inversion-templates-dir | TEXTUAL_INVERSION_TEMPLATES_DIR | textual_inversion_templates | Textual inversion范本的路径| -|--hypernetwork-dir | HYPERNETWORK_DIR | models/hypernetworks/ | Hypernetwork路径 | -|--localizations-dir | LOCALIZATIONS_DIR | localizations/ | 在地化翻译路径 | -|--styles-file | STYLES_FILE | styles.csv | 风格文件名 | -|--ui-config-file | UI_CONFIG_FILE | ui-config.json | UI设置档文件名 | -|--no-progressbar-hiding | None | False | 取消隐藏Gradio UI的进度条 (我们之所以将其隐藏,是因为在浏览器启动硬件加速的状况下,进度条会降低机器学习的性能) | -|--max-batch-count| MAX_BATCH_COUNT | 16 | UI的最大批量数值 | -|--ui-settings-file | UI_SETTINGS_FILE | config.json | UI设置值画面的文件名 | -|--allow-code | None | False | 允许在WebUI运行自订指令稿 | -|--share | None | False | 使用此参数在启动后会产生Gradio网址,使WebUI能从外部网络访问 | -|--listen | None | False | 以0.0.0.0主机名称启动Gradio,使其能回应连接请求 | -|--port | PORT | 7860 | 以给定的通信端口启动Gradio。1024以下的通信端口需要root权限。如果可用的话,缺省使用7860通信端口。 | -|--hide-ui-dir-config | None | False | 在WebUI隐藏设置档目录。| -|--freeze-settings | None | False | 停用编辑设置。 | -|--enable-insecure-extension-access | None | False | 无视其他选项,强制激活扩充功能页签。 | -|--gradio-debug | None | False | 使用 --debug选项启动Gradio | -|--gradio-auth | GRADIO_AUTH | None | 设置Gardio授权,例如"username:password",或是逗号分隔值形式"u1:p1,u2:p2,u3:p3" | -|--gradio-auth-path | GRADIO_AUTH_PATH | None | 设置Gardio授权文件路径。 例如 "/路径/" 再加上`--gradio-auth`的格式。 | -|--disable-console-progressbars | None | False | 不在终端机显示进度条。 | -|--enable-console-prompts | None | False | 在使用文生图和图生图的时候,于终端机印出提示词 | -|--api | None | False | 以API模式启动WebUI | -|--api-auth | API_AUTH | None | 设置API授权,例如"username:password",或是逗号分隔值形式"u1:p1,u2:p2,u3:p3" | -|--api-log | None | False | 激活所有API请求的纪录档 | -|--nowebui | None | False | 仅启动API, 不启动WebUI | -|--ui-debug-mode | None | False | 不加载模型,以更快启动WebUI | -|--device-id | DEVICE_ID | None | 选择要使用的CUDA设备 (例如在启动指令稿使用export CUDA_VISIBLE_DEVICES=0或1) | -|--administrator | None | False | 使用系统管理员权限 | -|--cors-allow-origins | CORS_ALLOW_ORIGINS | None | 允许跨来源资源共用,列表以逗号分隔,不可有空格 | -|--cors-allow-origins-regex | CORS_ALLOW_ORIGINS_REGEX | None | 允许跨来源资源共用,后面加上单一正规表达式 | -|--tls-keyfile | TLS_KEYFILE | None | 部份激活TLS,,需要配合--tls-certfile才能正常运作 | -|--tls-certfile | TLS_CERTFILE | None | 部份激活TLS,需要配合--tls-keyfile才能正常运作 | -|--server-name | SERVER_NAME | None | 设置服务器主机名称 | -|--skip-version-check | None | False | 不检查torch和xformers的版本 | -|--no-hashing | None | False | 停用计算存盘点模型的sha256哈希值,加快加载速度 | +|-h, --help |None |False | 显示此帮助消息并退出。 | +|--exit|||安装后终止程序。 | +|--data-dir|DATA_DIR|./|用户数据保存的路径。 | +|--config |CONFIG|configs/stable-diffusion/v1-inference.yaml |建构模型设置档的路径。 | +|--ckpt | CKPT |model.ckpt |Stable Diffusion模型的存盘点模型路径。一旦指定,该模型会加入至存盘点模型列表并加载。| +|--ckpt-dir | CKPT_DIR|None |存放Stable Diffusion模型存盘点模型的路径。 | +|--no-download-sd-model|None|False|即使找不到模型,也不自动下载SD1.5模型。 | +|--vae-dir|VAE_PATH|None | VAE的路径。 | +|--gfpgan-dir| GFPGAN_DIR|GFPGAN/ |GFPGAN路径 | +|--gfpgan-model| GFPGAN_MODEL |GFPGAN模型文件名 | +|--codeformer-models-path|CODEFORMER_MODELS_PATH|models/Codeformer/|Codeformer模型档的路径。 | +|--gfpgan-models-path|GFPGAN_MODELS_PATH|models/GFPGAN|GFPGAN模型档的路径。 | +|--esrgan-models-path|ESRGAN_MODELS_PATH|models/ESRGAN |ESRGAN模型档的路径。| +|--bsrgan-models-path|BSRGAN_MODELS_PATH|models/BSRGAN |BSRGAN模型档的路径。 | +|--realesrgan-models-path|REALESRGAN_MODELS_PATH|models/RealESRGAN|RealESRGAN模型档的路径。 | +|--scunet-models-path|SCUNET_MODELS_PATH|models/ScuNET| ScuNET模型档的路径。 | +|--swinir-models-path|SWINIR_MODELS_PATH|models/SwinIR| SwinIR和SwinIR v2模型档的路径。 | +|--ldsr-models-path|LDSR_MODELS_PATH|models/LDSR | 含有LDSR模型档的路径。 | +|--lora-dir|LORA_DIR|models/Lora|含有LoRA模型档的路径。| +|--clip-models-path|CLIP_MODELS_PATH|None|含有CLIP模型档的路径。 | +|--embeddings-dir|EMBEDDINGS_DIR|embeddings/ |Textual inversion的embeddings路径 (缺省: embeddings) | +|--textual-inversion-templates-dir|TEXTUAL_INVERSION_TEMPLATES_DIR|textual_inversion_templates|Textual inversion范本的路径| +|--hypernetwork-dir|HYPERNETWORK_DIR|models/hypernetworks/ |Hypernetwork路径 | +|--localizations-dir|LOCALIZATIONS_DIR|localizations/|在地化翻译路径 | +|--styles-file|STYLES_FILE|styles.csv | 风格文件名 | +|--ui-config-file|UI_CONFIG_FILE| ui-config.json | UI设置档文件名 | +|--no-progressbar-hiding|None|False |取消隐藏Gradio UI的进度条 (我们之所以将其隐藏,是因为在浏览器启动硬件加速的状况下,进度条会降低机器学习的性能) | +|--max-batch-count| MAX_BATCH_COUNT|16 |UI的最大批量数值 | +|--ui-settings-file|UI_SETTINGS_FILE|config.json | UI设置值画面的文件名 | +|--allow-code|None|False |允许在WebUI运行自订指令稿 | +|--share|None|False |使用此参数在启动后会产生Gradio网址,使WebUI能从外部网络访问 | +|--listen |None|False | 以0.0.0.0主机名称启动Gradio,使其能回应连接请求 | +|--port|PORT|7860 | 以给定的通信端口启动Gradio。1024以下的通信端口需要root权限。如果可用的话,缺省使用7860通信端口。 | +|--hide-ui-dir-config|None|False | 在WebUI隐藏设置档目录。| +|--freeze-settings|None|False|停用编辑设置。 | +|--enable-insecure-extension-access|None|False|无视其他选项,强制激活扩充功能页签。 | +|--gradio-debug|None|False | 使用 --debug选项启动Gradio | +|--gradio-auth|GRADIO_AUTH|None | 设置Gardio授权,例如"username:password",或是逗号分隔值形式"u1:p1,u2:p2,u3:p3" | +|--gradio-auth-path|GRADIO_AUTH_PATH|None| 设置Gardio授权文件路径。 例如 "/路径/" 再加上`--gradio-auth`的格式。 | +|--disable-console-progressbars|None|False | 不在终端机显示进度条。 | +|--enable-console-prompts|None|False | 在使用文生图和图生图的时候,于终端机印出提示词 | +|--api|None|False|以API模式启动WebUI | +|--api-auth|API_AUTH|None|设置API授权,例如"username:password",或是逗号分隔值形式"u1:p1,u2:p2,u3:p3" | +|--api-log|None|False|激活所有API请求的纪录档 | +|--nowebui|None|False|仅启动API, 不启动WebUI | +|--ui-debug-mode|None|False|不加载模型,以更快启动WebUI | +|--device-id|DEVICE_ID|None|选择要使用的CUDA设备 (例如在启动指令稿使用export CUDA_VISIBLE_DEVICES=0或1) | +|--administrator|None|False|使用系统管理员权限 | +|--cors-allow-origins|CORS_ALLOW_ORIGINS|None|允许跨来源资源共用,列表以逗号分隔,不可有空格 | +|--cors-allow-origins-regex|CORS_ALLOW_ORIGINS_REGEX|None|允许跨来源资源共用,后面加上单一正规表达式 | +|--tls-keyfile|TLS_KEYFILE|None|部份激活TLS,,需要配合--tls-certfile才能正常运作 | +|--tls-certfile|TLS_CERTFILE|None|部份激活TLS,需要配合--tls-keyfile才能正常运作 | +|--server-name|SERVER_NAME|None|设置服务器主机名称 | +|--skip-version-check|None|False|不检查torch和xformers的版本 | +|--no-hashing|None|False|停用计算存盘点模型的sha256哈希值,加快加载速度 | |--skip-version-check|None|False|不检查torch与xformers版本。| |--skip-version-check|None|False|不检查Python版本。| |--skip-torch-cuda-test|None|False|不检查CUDA是否正常运作。| @@ -121,51 +123,53 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram |--api-server-stop|None|False|允许通过API通信停止/重启/强制停止主程序。| |--timeout-keep-alive|int|30|设置uvicorn的timeout_keep_alive数值。| | **性能相关** | -|--xformers | None | False | 给cross attention layers激活xformers | -|--reinstall-xformers | None | False | 强制重装xformers,升级时很有用。但为避免不断重装,升级后将会移除。 | -|--force-enable-xformers | None | False | 强制给cross attention layers激活xformers ***此选项无法运作的话请勿回报bug*** | -|--xformers-flash-attention | None | False | 给xformers激活Flash Attention,提升再现能力 (仅支持SD2.x或以此为基础的模型)| -|--opt-split-attention | None | False | 强制激活Doggettx的cross-attention layer优化。有CUDA的系统缺省激活此选项。 | -|--opt-split-attention-invokeai | None | False | 强制激活InvokeAI的cross-attention layer优化。无CUDA的系统缺省激活此选项。 | -|--opt-split-attention-v1 | None | False | 激活旧版的split attention优化,防止占用全部可用的VRAM, | -|--opt-sub-quad-attention | None | False | 激活增进内存效率的sub-quadratic cross-attention layer优化| -|--sub-quad-q-chunk-size | SUB_QUAD_Q_CHUNK_SIZE | 1024 | sub-quadratic cross-attention layer优化使用的串行化区块大小| -|--sub-quad-kv-chunk-size | SUB_QUAD_KV_CHUNK_SIZE | None | sub-quadratic cross-attention layer优化使用的kv区块大小| -|--sub-quad-chunk-threshold | SUB_QUAD_CHUNK_THRESHOLD | None |sub-quadratic cross-attention layer优化过程中,区块化使用的VRAM阈值| -|--opt-channelslast | None | False | 激活4d tensors使用的alternative layout,或许可以加快推理速度 **仅适用搭载Tensor内核的Nvidia显卡(16xx系列以上)** | -|--disable-opt-split-attention | None | False | 强制停用cross-attention layer的优化 | -|--disable-nan-check | None | False | 不检查生成图像/潜在空间是否有nan。在CI模式无使用存盘点模型的时候很有用。| -|--use-cpu | {all, sd, interrogate, gfpgan, bsrgan, esrgan, scunet, codeformer} | None | 让部份模块使用CPU作为PyTorch的设备 | -|--no-half | None | False | 不将模型转换为半精度浮点数 | -|--precision | {full,autocast} | autocast | 使用此精度评估 | -|--no-half-vae | None | False | 不将VAE模型转换为半精度浮点数 | -|--upcast-sampling | None | False | 向上采样。搭配 --no-half使用则无效。生成的结果与使用--no-half参数相近,效率更高,使用更少内存。| -|--medvram | None | False | 激活Stable Diffusion模型优化,牺牲速度,换取较小的VRAM占用。 | -|--lowvram | None | False | 激活Stable Diffusion模型优化,大幅牺牲速度,换取更小的VRAM占用。 | -|--lowram | None | False | 将Stable Diffusion存盘点模型的权重加载至VRAM,而非RAM | +|--xformers|None|False | 给cross attention layers激活xformers | +|--reinstall-xformers|None|False | 强制重装xformers,升级时很有用。但为避免不断重装,升级后将会移除。 | +|--force-enable-xformers | None|False | 强制给cross attention layers激活xformers ***此选项无法运作的话请勿回报bug*** | +|--xformers-flash-attention|None|False|给xformers激活Flash Attention,提升再现能力 (仅支持SD2.x或以此为基础的模型)| +|--opt-split-attention|None|False | 强制激活Doggettx的cross-attention layer优化。有CUDA的系统缺省激活此选项。 | +|--opt-split-attention-invokeai|None|False | 强制激活InvokeAI的cross-attention layer优化。无CUDA的系统缺省激活此选项。 | +|--opt-split-attention-v1|None|False | 激活旧版的split attention优化,防止占用全部可用的VRAM, | +|--opt-sub-quad-attention|None|False|激活增进内存效率的sub-quadratic cross-attention layer优化| +|--sub-quad-q-chunk-size|SUB_QUAD_Q_CHUNK_SIZE|1024|sub-quadratic cross-attention layer优化使用的串行化区块大小| +|--sub-quad-kv-chunk-size|SUB_QUAD_KV_CHUNK_SIZE|None|sub-quadratic cross-attention layer优化使用的kv区块大小| +|--sub-quad-chunk-threshold|SUB_QUAD_CHUNK_THRESHOLD|None |sub-quadratic cross-attention layer优化过程中,区块化使用的VRAM阈值| +|--opt-channelslast|None|False | 激活4d tensors使用的alternative layout,或许可以加快推理速度 **仅适用搭载Tensor内核的Nvidia显卡(16xx系列以上)** | +|--disable-opt-split-attention|None|False | 强制停用cross-attention layer的优化 | +|--disable-nan-check|None|False|不检查生成图像/潜在空间是否有nan。在CI模式无使用存盘点模型的时候很有用。| +|--use-cpu|{all, sd, interrogate, gfpgan, bsrgan, esrgan, scunet, codeformer}|None|让部份模块使用CPU作为PyTorch的设备 | +|--no-half |None|False | 不将模型转换为半精度浮点数 | +|--precision|{full,autocast}|autocast |使用此精度评估 | +|--no-half-vae|None|False |不将VAE模型转换为半精度浮点数 | +|--upcast-sampling|None|False|向上采样。搭配 --no-half使用则无效。生成的结果与使用--no-half参数相近,效率更高,使用更少内存。| +|--medvram |None|False |激活Stable Diffusion模型优化,牺牲速度,换取较小的VRAM占用。 | +|--lowvram |None|False |激活Stable Diffusion模型优化,大幅牺牲速度,换取更小的VRAM占用。 | +|--lowram |None|False |将Stable Diffusion存盘点模型的权重加载至VRAM,而非RAM | |--disable-model-loading-ram-optimization|None|False|停用模型加载时降低RAM占用的优化。| | **功能** | -|--autolaunch | None | False | 启动WebUI后自动打开系统缺省的浏览器 | -|--theme | None | Unset | 使用指定主题启动WebUI (light或dark),无指定则使用浏览器缺省主题。 | -|--use-textbox-seed | None | False | 在WebUI的种子字段使用textbox (没有上下,但可以输入长的种子码) | -|--disable-safe-unpickle | None | False | 不检查PyTorch模型是否有恶意代码 | -|--ngrok | NGROK | None | Ngrok授权权杖, --share参数的替代品。 | -|--ngrok-region | NGROK_REGION | us | 选择启动Ngrok的区域 | +|--autolaunch|None|False | 启动WebUI后自动打开系统缺省的浏览器 | +|--theme|None|Unset | 使用指定主题启动WebUI (light或dark),无指定则使用浏览器缺省主题。 | +|--use-textbox-seed|None|False | 在WebUI的种子字段使用textbox (没有上下,但可以输入长的种子码) | +|--disable-safe-unpickle|None|False | 不检查PyTorch模型是否有恶意代码 | +|--ngrok|NGROK|None |Ngrok授权权杖, --share参数的替代品。 | +|--ngrok-region|NGROK_REGION|us |选择启动Ngrok的区域 | |--update-check|None|None|启动时检查有无新版本。| |--update-all-extensions|None|None|在启动WebUI的时候自动更新所有扩充功能。| |--reinstall-xformers|None|False|强制重新安装xformers,适用于更新程序之后运行。更新完之后记得移除此参数。| |--reinstall-torch|None|False|强制重新安装touch,适用于更新程序之后运行。更新完之后记得移除此参数。| |--tests|TESTS|False|运行功能测试,确认WebUI正常运作。| -|--no-tests| None | False | 即使有`--test`参数也不要运行功能测试。 | -|--dump-sysinfo | None | False | 倾印系统消息并退出程序(不包括扩充功能) | +|--no-tests| None|False|即使有`--test`参数也不要运行功能测试。 | +|--dump-sysinfo | None|False|倾印系统消息并退出程序(不包括扩充功能) | |--disable-all-extensions|None|False|停用所有扩充功能,包含内置的扩充功能。| |--disable-extra-extensions|None|False|停用所有扩充功能。| | **已经无效的选项** | -|--show-negative-prompt | None | False | 无作用 | -|--deepdanbooru | None | False |无作用 | -|--unload-gfpgan | None | False | 无作用 | -|--gradio-img2img-tool | GRADIO_IMG2IMG_TOOL | None | 无作用 | -|--gradio-inpaint-tool | GRADIO_INPAINT_TOOL | None | 无作用 | -|--gradio-queue | None | False | 无作用 | -|--add-stop-route | None | False | 无作用 | -|--always-batch-cond-uncond | None | False | 无作用 | +|--show-negative-prompt|None|False | 无作用 | +|--deepdanbooru|None|False |无作用 | +|--unload-gfpgan|None|False |无作用 | +|--gradio-img2img-tool|GRADIO_IMG2IMG_TOOL|None|无作用 | +|--gradio-inpaint-tool|GRADIO_INPAINT_TOOL|None|无作用 | +|--gradio-queue|None|False|无作用 | +|--add-stop-route|None|False|无作用 | +|--always-batch-cond-uncond|None|False|无作用 | + +
diff --git a/content.zh-cn/installation/deploy-to-paperspace.md b/content.zh-cn/installation/deploy-to-paperspace.md deleted file mode 100755 index 35da9b5..0000000 --- a/content.zh-cn/installation/deploy-to-paperspace.md +++ /dev/null @@ -1,47 +0,0 @@ ---- -title: "☁️ 部署至Paperspace" -weight: 12 ---- - -Paperspace是一间位于美国的云端机器学习服务供应商,提供免费GPU跑机器学习项目。 - -免费版本的保存空间5GB,只能使用有限制的运行个体,一次只能运行一个笔记本,6小时后机器自动关闭。有时候免费GPU需要排队。 - -值得一提的是Paperspace官方有维护专门跑Stable Diffusion WebUI的笔记本,点一下就可以使用,并且速度很快。代码很清晰,可按照需求修改。 - - -1. 至[Paperspace](https://www.paperspace.com/)注册帐号。注册需要提供电话号码。 - -2. 进入[gradient-ai/stable-diffusion](https://console.paperspace.com/github/gradient-ai/stable-diffusion?machine=Free-GPU&ref=blog.paperspace.com)页面,点击Start Machine - -![](../../../images/deploy-to-paperspace-1.webp) - -3. 这个皮肤就是笔记本的页面了。点击右上角的Run all后,笔记本会安装Stable Diffusion WebUI所需的依赖,并下载Stable Diffusion v2.1的模型。此阶段所下载的文件不会计入帐户的保存空间。 - -![](../../../images/deploy-to-paperspace-2.webp) - - -4. 之后点击Use the Stable Diffusion Web UI这个保存格终端机显示的Gradio网址 - -![](../../../images/deploy-to-paperspace-3.webp) - - -5. 即可使用WebUI。 - -![](../../../images/deploy-to-paperspace-4.webp) - -6. 使用完毕后记得按左上角的Stop Machine关机。 - - -此笔记本只有安装最小功能。扩充功能和额外模型要自行修正笔记本程序下载。 - -下载自订模型可以加在拷贝Stable Diffusion WebUI的保存库之后 -```python -!git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui -!wget -nc -P stable-diffusion-webui/models/Stable-diffusion https://huggingface.co/andite/anything-v4.0/resolve/main/anything-v4.5-pruned.safetensors -``` - -参照[命令行参数](../installation/command-line-arguments-and-settings/),找到启动WebUI这行,加入`--enable-insecure-extension-access`让你可从网页版的Extension页面下载扩充。 -```python -!python launch.py --share --config ~/../datasets/stable-diffusion-classic-v2/768-v-ema.yaml --ckpt ~/../datasets/stable-diffusion-classic-v2/768-v-ema.ckpt --enable-insecure-extension-access -``` diff --git a/content.zh-cn/installation/download-models.md b/content.zh-cn/installation/download-models.md index a8ae3dc..b5e4a7b 100644 --- a/content.zh-cn/installation/download-models.md +++ b/content.zh-cn/installation/download-models.md @@ -20,9 +20,8 @@ Stable Diffusion WebUI在初次启动时,若放模型的文件夹(`models/Stab 另一个是[Civitai](https://civitai.com/),专门用来分享Stable Diffusion相关的资源,特色是模型都有示范缩略图,用户也可以分享彼此使用的提示词,以及分享作品。 -{{< hint danger >}} -Civitai可能包含未成年不宜的图像。 -{{< /hint >}} +Civitai可能包含未成年不宜的图像。 + ## 写实风模型 @@ -38,21 +37,19 @@ Civitai可能包含未成年不宜的图像。 [Deliberate](https://civitai.com/models/4823/deliberate):基于SD-1.5模型,适合生成精致写实风格的人物、动物、自然风景。 -[Realistic Vision v1.4](https://civitai.com/models/4201/realistic-vision-v13):写实风人物与动物模型。 +[Realistic Vision v6.0](https://civitai.com/models/4201/realistic-vision-v60-b1):写实风人物与动物模型。 ## 动漫风模型 -[Anything万象熔炉 v4.5](https://huggingface.co/andite/anything-v4.0)适合画动漫图,作者宣称不需要打一堆提示词也能出漂亮的图。 - -[Waifu Diffusion v1.4](https://huggingface.co/hakurei/waifu-diffusion-v1-4)是纯粹使用Danbooru图库训练而成,适合画动漫图。 - -[Hentai Diffusion](https://github.com/Delcos/Hentai-Diffusion)适合画动漫图,模型已使用大量负向提示词训练过以排除不良结果,另提供embeddings方便绘图时使用。 +[Anything万象熔炉 v5](https://civitai.com/models/9409/anything-or)适合画动漫图,作者宣称不需要打一堆提示词也能出漂亮的图。 [DreamShaper](https://civitai.com/models/4384/dreamshaper)是基于SD-1.5模型,生成精细动漫人物与油画风格的模型。 [OrangeMix3](https://huggingface.co/WarriorMama777/OrangeMixs),混合多种风格的动漫绘图模型,偏写实。 +[Waifu Diffusion v1.4](https://huggingface.co/hakurei/waifu-diffusion-v1-4)是纯粹使用Danbooru图库训练而成,适合画动漫图。 + # 2. 如何下载模型 @@ -64,7 +61,6 @@ HuggingFace为保存库 → Files and versions,选择下载扩展名为`.ckpt` Civitai则是点击Download按钮下载模型。 - 如果要用指令下载,可以用wget或git,例如下载`Anything V4.5`的模型到`stable-diffusion-webui/models/Stable-diffusion`文件夹: ```bash wget -nc -P stable-diffusion-webui/models/Stable-diffusion https://huggingface.co/andite/anything-v4.0/resolve/main/anything-v4.5-pruned.safetensors diff --git a/content.zh-cn/installation/system-requirements.md b/content.zh-cn/installation/system-requirements.md index d39da98..48566f0 100644 --- a/content.zh-cn/installation/system-requirements.md +++ b/content.zh-cn/installation/system-requirements.md @@ -53,15 +53,13 @@ weight: 1 - 要生成513x512的图,显卡VRAM在4GB以下的很容易遇到内存不足的问题,并且生图速度很慢。 -- 尽管可以[用纯CPU跑Stable Diffusion WebUI](https://ivonblog.com/posts/stable-diffusion-running-on-cpu/),但是速度会非常慢。一张显卡30秒就能算好的图,CPU要算10分钟。因此用独立显卡跑Stable Diffsuion仍是比较好的选择。 +- 尽管可以[用纯CPU跑Stable Diffusion WebUI](https://ivonblog.com/posts/stable-diffusion-running-on-cpu/),但是速度会非常慢。一张显卡30秒就能算好的图,CPU要算10分钟。因此用独立显卡跑Stable Diffusion仍是比较好的选择。 - 硬件不够力的话,用云端服务跑也是可以,例如Google Colab。 # 2. 操作系统需求 -## GNU/Linux - 测试可以跑的Linux发行版: - Debian 12 diff --git a/content.zh-cn/tutorials/text-to-image-controlnet-latent-couple.md b/content.zh-cn/tutorials/text-to-image-controlnet-latent-couple.md index 1eebe52..1087319 100644 --- a/content.zh-cn/tutorials/text-to-image-controlnet-latent-couple.md +++ b/content.zh-cn/tutorials/text-to-image-controlnet-latent-couple.md @@ -3,7 +3,7 @@ title: "用文生图 + ControlNet + Latent Couple生成指定位置的多个 weight: 1 --- -请先阅读[文生图用法](../features/text-to-image/),并安装[ControlNet](../extensions/controlnet/)、[Latent Couple](../extensions/latent-couple/)、[PoseX](../extensions/posex/)扩充功能。 +请先阅读[文生图用法](../features/text-to-image/),并安装ControlNet、Latent Couple、PoseX扩充功能。 ![](../../../images/text-to-image-controlnet-latent-couple-7.webp) diff --git a/content.zh-tw/_index.md b/content.zh-tw/_index.md index 055cea6..2d683fd 100644 --- a/content.zh-tw/_index.md +++ b/content.zh-tw/_index.md @@ -4,7 +4,7 @@ bookFlatSection: true title: "歡迎" --- -這本非官方的中文使用手冊是針對AUTOMATIC1111開發的Stable Diffusion WebUI所撰寫,對應版本為**v1.6.0** +這本非官方的中文使用手冊是針對AUTOMATIC1111開發的Stable Diffusion WebUI所撰寫,對應版本為**v1.6.0**。最後更新日期如網頁底部所示。 內容有:介紹如何將Stable Diffusion WebUI安裝至您的電腦,以及講解相關AI繪圖功能用法,最後再介紹模型訓練方式。 diff --git a/content.zh-tw/extensions/_index.md b/content.zh-tw/extensions/_index.md index 95e8c87..de89509 100644 --- a/content.zh-tw/extensions/_index.md +++ b/content.zh-tw/extensions/_index.md @@ -6,6 +6,8 @@ bookCollapseSection: true Stable Diffusion WebUI除了基本的文生圖、圖生圖、內補繪製之外,還能安裝擴充功能(Extensions,或稱「插件」),在繪圖的時候整合其他AI技術,或是簡化操作,生成更好的作品。 -這一章介紹設計給Stable Diffusion WebUI的擴充功能用法,並附上一些範例。包含界面中文化、控制動作的ControlNet、生成多個人物的Latent Couple等等。 +例如:界面中文化、控制動作的ControlNet、生成多個人物的Latent Couple的擴充功能等等。 + +因為擴充功能實在太多,這一章僅介紹Stable Diffusion WebUI的擴充功能裝法,各個擴充功能的細節部份不多談。 針對Stable Diffusion WebUI設計的擴充功能非常多,本章列表不定時更新。 diff --git a/content.zh-tw/extensions/a1111-sd-webui-tagcomplete.md b/content.zh-tw/extensions/a1111-sd-webui-tagcomplete.md deleted file mode 100644 index 3373825..0000000 --- a/content.zh-tw/extensions/a1111-sd-webui-tagcomplete.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "提示詞自動補完 Booru tag autocompletion" -weight: 5 ---- - -- 擴充功能來源:[DominikDoom/a1111-sd-webui-tagcomplete](https://github.com/DominikDoom/a1111-sd-webui-tagcomplete) - -輸入提示詞的時候提示相關Danbooru上的標籤,畫動漫圖的時候很有用,還可以安裝中文標籤翻譯。 - -![](../../images/a1111-sd-webui-tagcomplete-1.webp) - -進入Extensions頁面,點選`Install from URL`,輸入網址`https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git`,按`Install`,之後重啟WebUI。 - -如何安裝標籤中文翻譯:到[byzod/a1111-sd-webui-tagcomplete-CN](https://github.com/byzod/a1111-sd-webui-tagcomplete-CN)儲存庫,點選綠色的Code → Download ZIP,解壓縮,得到tags資料夾,下面有`Tags-zh-full-pack.csv`和`config.json`這二個檔案。將這二個檔案移動到`stable-diffusion-webui/extensions/a1111-sd-webui-tagcomplete/tags`資料夾。 diff --git a/content.zh-tw/extensions/auto-sd-paint-ext.md b/content.zh-tw/extensions/auto-sd-paint-ext.md deleted file mode 100644 index e6b17f8..0000000 --- a/content.zh-tw/extensions/auto-sd-paint-ext.md +++ /dev/null @@ -1,56 +0,0 @@ ---- -title: "繪圖軟體Krita整合 auto-sd-paint-ext" -weight: 10 ---- - -- 擴充功能來源:[Interpause/auto-sd-paint-ext](https://github.com/Interpause/auto-sd-paint-ext) - -這是將Stable Diffusion的Inpainti整合至繪圖軟體Krita的擴充功能,可在選取圖層範圍後調用Stable Diffusion填充內容,支援不規則選取。 - -1. 進入Extension頁面,點選Install from URL:`https://github.com/Interpause/auto-sd-paint-ext.git`,關閉SD WebUI。 - -2. 編輯SD WebUI的啟動引數,加入`--api`啟動SD WebUI。進入網頁界面會看到多了`auto-sd-paint-ext Guide/Panel`頁面 - -![](../../images/auto-sd-paint-ext-1.webp) - -3. 安裝自由軟體[Krita](https://krita.org/) - -4. 啟動Krita,點選上方的設定 → 管理資源,點選「開啟資源資料夾」 - -![](../../images/auto-sd-paint-ext-2.webp) - -5. 複製`pykrita`資料夾的路徑 - -![](../../images/auto-sd-paint-ext-3.webp) - -6. 回到SD WebUI,貼上pykrita資料夾的路徑,下面即會產生建立符號連結的指令。開啟終端機,按照你的作業系統執行指令。 - -![](../../images/auto-sd-paint-ext-4.webp) - -7. 重新啟動Krita。點選設定 → 設定Krita,點選左邊的「Python外掛程式管理器」,再勾選右邊的Stable Diffusion Plugin - -![](../../images/auto-sd-paint-ext-5.webp) - -8. 重新啟動Krita,編輯圖片時左邊即會多出一堆面板。這些面板可從設定 → 工具面板叫出來 - -![](../../images/auto-sd-paint-ext-6.webp) - -9. 以Inpaint的功能為例子,讓我們先把無關的面板都關掉,只留下"Inpaint" - -![](../../images/auto-sd-paint-ext-7.webp) - -10. 點選右下角新增顏料圖層 - -![](../../images/auto-sd-paint-ext-8.webp) - -11. 在此圖層用鉛筆工具,將要填充的地方塗黑 - -![](../../images/auto-sd-paint-ext-9.webp) - -12. 於左邊的面板填入提示詞,點選Start Inpaint,開始繪圖。進度條可從終端機查看。 - -![](../../images/auto-sd-paint-ext-10.webp) - -13. SD WebUI運算完成後結果即會顯示在Krita的圖層上。 - -![](../../images/auto-sd-paint-ext-11.webp) diff --git a/content.zh-tw/extensions/controlnet.md b/content.zh-tw/extensions/controlnet.md deleted file mode 100644 index 01fc0bb..0000000 --- a/content.zh-tw/extensions/controlnet.md +++ /dev/null @@ -1,223 +0,0 @@ ---- -title: "ControlNet 按照骨架動作繪圖" -weight: 3 ---- - -- 擴充功能來源:[Mikubill/sd-webui-controlnet](https://github.com/Mikubill/sd-webui-controlnet) -- ControlNet的儲存庫:[lllyasviel/ControlNet](https://github.com/lllyasviel/ControlNet) - -目前版本:ControlNet v1.1 - -ControlNet是通過加入額外條件來控制擴散模型的神經網路結構,它可以讓AI參考給定圖片的動作/線條/景深,更精準的生成圖片。 - -跟內建的「圖生圖」技術比起來,ControlNet的效果更好,能讓AI以指定動作生圖;再搭配3D建模作為輔助,能緩解單純用文生圖手腳、臉部表情畫不好的問題。 - -ControlNet的用法還有:上傳人體骨架線條,ControlNet就能按骨架的動作生成完稿的人物圖片。或是上傳素色的3D建模,讓ControlNet彩現成為室內佈置家具。 - -Lvmin Zhang是ControlNet原始程式的開發者,Mikubill則是開發擴充功能,讓我們可以在Stable Diffusion WebUI用ControlNet生圖。 - - -# 1. 安裝ControlNet與下載模型 - -1. 切換至Extensions頁面,點選`Install From URL`,URL輸入`https://github.com/Mikubill/sd-webui-controlnet.git`,按下Install,接著從終端機按CTRL+C,關閉Stable Diffusion WebUI。 - -2. 到[lllyasviel/ControlNet v1.1](https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main)下載以`.pth`結尾的模型檔,全部檔案加起來約18GB。`.yaml`檔不需要下載。目前v1.1是測試版,之後會合併回[主儲存庫](https://huggingface.co/lllyasviel/ControlNet)。 - -![](../../images/controlnet-1.webp) - -3. 將這些模型檔放到`stable-diffusion-webui/extensions/sd-webui-controlnet/models`資料夾 - -4. 重新啟動Stable Diffusion WebUI。 - - -# 2. 各個模型的用途 - -ControlNet現有以下模型,您可以按照需求只下載需要的模型。 - -## Anime Lineart - -偵測線條,生成的圖片亦會保留原始的線條,適合處理動漫圖像 - -## Canny - -偵測圖片邊緣,比較模糊,不如Scribbles完整。 - -## Depth - -偵測輸入圖片的深度圖(depth map)。 - -## Illumination - -偵測輸入圖片的光源與照明效果。 - -## Inpaint - -功能類似「內補繪製」,使用50%隨機遮罩+50%隨機光流遮罩訓練而成。 - -## Instruct Pix2Pix - -模型檔名為`ip2p`,類似「圖生圖」,但是使用訓練50%的指示(instruction)提示詞和50%的敘述(description)提示詞訓練而成。因為是ControlNet,使用此模型時不需要調整CFG Scale。 - -根據原作者的說法,此模型在下「使其成為X」的提示詞所生成的圖,效果比「使Y成為X」要好。 - -> Also, it seems that instructions like "make it into X" works better than "make Y into X". - -## Lineart - -偵測線條,適合處理線稿,生成的圖片亦會保留原始的線條。 - -## M-LSD - -偵測輸入圖片的直線。 - -## Normalbae - -## Openpose - -使用OpenPose技術偵測輸入圖片人物的動作,不一定會保留線條。 - -## Scribbles - -偵測線條,偵測到的線條品質介於Soft Edge和Lineart之間。 - -## Segmentation - -模型檔名為`seg`,將偵測的圖片物件切成一個一個色塊處理,例如房子一個色塊,後面的天空一個色塊。 - -## Shuffle - -把輸入圖片的概念轉移到生成的圖片。 - -作者給的例子:輸入灰色裝甲圖片,生成的鋼鐵人盔甲也會是灰色的。 - -## Soft Edge - -偵測圖片邊緣,效果較為柔和,像用炭筆塗過。 - -## Tile - -輸入圖片,選取一個區域,使其變清晰的模型。 - - -# 3. 參數解說 - -在安裝之後,WebUI的界面會多出"ControlNet"的按鈕。 - -![](../../images/controlnet-2.webp) - -點開會看到以下選項: - -## Single Image - -上傳單張圖片。 - -## Batch - -處理多張圖片。Input Directory填寫的是待處理的圖片所在路徑。 - -## Open New Canvas - -清除圖片,開一張新畫布。 - -## 使用電腦鏡頭 - -在Open New Canvas的選項下面有四個按鈕。 - -![](../../images/controlnet-3.webp) - -由左到右:新增畫布;使用電腦鏡頭拍照上傳;未知;還原上次算圖設定。 - -## Enable - -在生圖時啟用ControlNet,必選。 - -## Low VRAM - -生圖使用ControlNet會吃更多VRAM,此選項可降低VRAM使用量。 - -## Pixel Perfect - -由ControlNet自動決定Preprocessor解析度。 - -## Allow Preview - -允許預覽生圖結果,會顯示該模型偵測的線條或骨架。 - -## Preprocessor 預處理器 - -通常Preprocessor和Model二者要一致。 - -Preprocessor是先將上傳的圖片處理過一輪,例如`Scribbles`會將彩色圖片轉成線稿。若圖片的線條是白底黑線,且不是用WebUI畫布所繪製,則Preprocessor必須選取"Invert"。 - -如果你的圖片不需要處理,Preprocessor可留空。 - -## Run Preprocessor (爆炸圖示) - -執行預處理器,預先處理圖片。 - - -## Model 模型 - -設定生圖的時候使用哪個ControlNet模型。 - -## Control Weight - -ControlNet在生圖時的權重。 - -## Starting Control Step - -要讓ControlNet開始參與生圖的步數。 - -## Ending Control Step - -讓ControlNet結束參與生圖的步數。 - -## Preprocessor Resolution - -預處理器的解析度。 - -如果不知道要設多少,請勾選"Pixel Perfect"讓ControlNet自動為您決定。 - -## Control Mode (Guess Mode) - -控制生圖時,要AI平衡二者(Balanced),或是偏重你的提示詞(My prompt is more important),還是偏重ControlNet (ControlNet is more important)。 - -## [Loopback] Automatically send generated images to this ControlNet unit - -自動將生成的圖片傳回此ControlNet運算單元(unit)。 - -是的,如果您的電腦VRAM夠大,您可以在Settings → ControlNet啟用多個ControlNet運算單元。 - -## Multi ControlNet - -一個ControlNet效果不夠,您有沒有想過啟用第二、第三個ControlNet!? - -預設安裝後只會使用一個ControlNet,但是疊加多個有助於生成更好的圖片。比方說同時使用Open Pose和Depth生成人體,並防止手畫歪。 - -要啟用Multi ControlNet,請進入Settings → ControlNet,拖動拉桿,設定要啟用的ControlNet數量。 - - - - -# 4. 實際操作 - -此處以改繪一個人物姿勢為例子。 - -1. 在安裝之後,WebUI的界面會多出"ControlNet"的按鈕。 - -![](../../images/controlnet-5.webp) - -2. 進入文生圖的頁面填入提示詞,接著點開下面ControlNet的界面,勾選`Enabled`啟用ControlNet,上傳圖片。勾選LowVRAM可降低VRAM佔用。 - -![](../../images/controlnet-6.webp) - -3. 接著選取要使用的Preprocessor和Model,二者要一致。 - -Preprocessor是先將上傳的圖片處理過一輪,例如`Scribbles`會將彩色圖片轉成線稿。如果你的圖片不需要處理,則Preprocessor可留空。 - -![](../../images/controlnet-7.webp) - -4. 以Scribbles為例,選取後點選右上角Generate即會生成出結果,並附上偵測到的線條。如下圖所示,右邊貞德的姿勢確實跟左邊的02十分類似。 - -![](../../images/controlnet-8.webp) - diff --git a/content.zh-tw/extensions/easynegative.md b/content.zh-tw/extensions/easynegative.md deleted file mode 100644 index 45e6d48..0000000 --- a/content.zh-tw/extensions/easynegative.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "簡易負向提示詞 EasyNegative" -weight: 4 ---- - -- 擴充功能來源:[gsdf/EasyNegative](https://huggingface.co/datasets/gsdf/EasyNegative) - -用這個取代冗長的負向提示詞吧,反過來利用Embedding的特性,告訴AI不要畫什麼。 - -到[作者的儲存庫](https://huggingface.co/datasets/gsdf/EasyNegative)下載`EasyNegative.safetensors`,將它放到`stable-diffuison-webui`資料夾下的`embeddings`,重啟WebUI。 - -之後在WebUI的負向提示詞欄位填入`EasyNegative`即會生效,可有效替代冗長的負向提示詞。不過單獨使用效果可能還是不好,你還是需要額外準備一些負向提示詞。 - -![](../../images/easynegative-1.webp) diff --git a/content.zh-tw/extensions/latent-couple.md b/content.zh-tw/extensions/latent-couple.md deleted file mode 100644 index 60de6b4..0000000 --- a/content.zh-tw/extensions/latent-couple.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "生成多個人物 Latent Couple" -weight: 8 ---- - -- 擴充功能來源:[ashen-sensored/stable-diffusion-webui-two-shot](https://github.com/ashen-sensored/stable-diffusion-webui-two-shot) - -Latent Couple會分割繪圖時的提示詞,這樣就可以生成多重人物/物件,並精確指定位置顏色了。 - -像是先用PoseX拉好骨架,再使用Latent Couple標出人物的概略位置。 - -於Extensions頁面的Install from URL輸入以下網址安裝: `https://github.com/ashen-sensored/stable-diffusion-webui-two-shot.git`,再點選Reload WebUI。 - -安裝後於文生圖的界面即會看到Latent Couple的選項。 diff --git a/content.zh-tw/extensions/multidiffusion-upscaler-for-automatic1111.md b/content.zh-tw/extensions/multidiffusion-upscaler-for-automatic1111.md deleted file mode 100644 index 579e701..0000000 --- a/content.zh-tw/extensions/multidiffusion-upscaler-for-automatic1111.md +++ /dev/null @@ -1,16 +0,0 @@ ---- -title: "低VRAM也能生大圖 Tiled Diffusion" -weight: 5 ---- - -- 擴充功能來源:[pkuliyi2015/multidiffusion-upscaler-for-automatic1111](https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111) - -這個擴充功能讓VRAM偏低(< 6GB)的電腦也能生成1080x1080以上解析度的大型圖片,甚至不需要用`--lowvram`和`--medvram`引數來啟動Stable Diffusion WebUI了。 - -安裝方法:進入Extensions頁面,點選Install from URL,輸入`https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111.git`後按Install,重啟WebUI。 - -在文生圖的頁面左下角會看到Tiled Diffusion和Tiled VAE,將二者的Enabled勾選即為啟用,生圖時維持預設值即可。 - -![](../../images/multidiffusion-upscaler-for-automatic1111-1.webp) - -另提供Region Prompt Control,可以控制每個tile的提示詞。 diff --git a/content.zh-tw/extensions/openpose-for-blender.md b/content.zh-tw/extensions/openpose-for-blender.md deleted file mode 100644 index 231df84..0000000 --- a/content.zh-tw/extensions/openpose-for-blender.md +++ /dev/null @@ -1,12 +0,0 @@ ---- -title: "3D人體+手部+臉部模型 Openpose for blender" -weight: 12 ---- - -- 擴充功能來源:[Character bones that look like Openpose for blender _ Ver_8.1 Depth+Canny+Landmark+MediaPipeFace+finger](https://toyxyz.gumroad.com/l/ciojz) - -設計給Blender使用的3D人體模型,整合OpenPose和手部、臉部的模型,在Blender內擺好姿勢後輸出圖片,即可搭配ControlNet生圖。 - -*需要安裝Blender 3.0以上版本。 - -作者開放自由捐款,不想花錢的話金額填0即可下載。 diff --git a/content.zh-tw/extensions/posex.md b/content.zh-tw/extensions/posex.md deleted file mode 100644 index 5feb548..0000000 --- a/content.zh-tw/extensions/posex.md +++ /dev/null @@ -1,32 +0,0 @@ ---- -title: "骨架人偶 PoseX" -weight: 6 ---- - -- 擴充功能來源:[hnmr293/posex](https://github.com/hnmr293/posex) - -*需要先安裝ControlNet才能使用這個擴充功能。 - -PoseX是可以在Stable Diffuison WebUI直接拉人物骨架,再配合ControlNet生成姿勢的擴充功能。 - -類似的擴充功能為[OpenPose Editor](https://github.com/fkunn1326/openpose-editor),差別在於PoseX可以旋轉放大。 - -1. 切換至Extensions頁面,點選Install From URL,URL輸入`https://github.com/hnmr293/posex.git`,按下Install。接著重啟WebUI。 - -![](../../images/posex-1.webp) - -2. 開啟文生圖的頁面,點選右下角PoseX,點選Send this image to ControlNet - -![](../../images/posex-2.webp) - -3. 在下面的ControlNet,點選Enabled,preprocessor選取`none`,model選`openpose`,不需要上傳圖片。 - -![](../../images/posex-3.webp) - -4. 回到上面的PoseX,調整人物姿勢。左鍵點選移動,滾輪放大縮小,對模型左鍵點二下即可用右鍵移動單個骨架。 - -![](../../images/posex-4.webp) - -5. 填入正負向提示詞,即會按照PoseX的姿勢來生成圖片。 - -![](../../images/posex-5.webp) diff --git a/content.zh-tw/extensions/sd-extension-system-info.md b/content.zh-tw/extensions/sd-extension-system-info.md deleted file mode 100644 index e4101aa..0000000 --- a/content.zh-tw/extensions/sd-extension-system-info.md +++ /dev/null @@ -1,14 +0,0 @@ ---- -title: "即時監控系統資訊 System Info" -weight: 5 ---- - -- 擴充功能來源:[vladmandic/sd-extension-system-info](https://github.com/vladmandic/sd-extension-system-info) - -此擴充功能給Stable Diffusion WebUI新增一個監控系統資訊的頁面,顯示主程式版本、系統資訊、GPU、RAM、使用的模型、使用的命令列引數等等。 - -![](../../images/sd-extension-system-info-1.webp) - -點選下面的Run benchmark可以跑分測試系統性能,有助於評估主程式更新後性能是否有變化。點選Submit results將結果回報到[Vladimir Mandic的網站](https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html)。 - -![](../../images/sd-extension-system-info-2.webp) diff --git a/content.zh-tw/extensions/sd-webui-depth-lib.md b/content.zh-tw/extensions/sd-webui-depth-lib.md deleted file mode 100644 index 0971123..0000000 --- a/content.zh-tw/extensions/sd-webui-depth-lib.md +++ /dev/null @@ -1,34 +0,0 @@ ---- -title: "手部模型 Depth map library and poser" -weight: 7 ---- - -- 擴充功能來源:[jexom/sd-webui-depth-lib](https://github.com/jexom/sd-webui-depth-lib) - -此擴充功能加入手部模型,可配合PoseX的骨架使用,改善AI手部畫不好的問題。 - -於Extensions頁面的Install from URL輸入以下網址安裝: `https://github.com/jexom/sd-webui-depth-lib`,重啟WebUI。 - -這裡示範搭配PoseX畫手的用法,請記得啟用Multi ControlNet。 - -1. 於文生圖頁面點開PoseX,設定`Target ControlNet number`為1,擺好骨架,並設定。然後點選Save PNG,得到一張png圖檔 - -![](../../images/sd-webui-depth-lib-1.webp) - -2. 切換至Depth map library頁面,點選Add Background Image,上傳PoseX骨架的圖檔。 - -![](../../images/sd-webui-depth-lib-2.webp) - -3. 再點選左邊各個手的模型,加到右邊的圖片,將二隻手擺出來。再點Send to Img2img。 - -![](../../images/sd-webui-depth-lib-3.webp) - -4. 回到文生圖頁面,點開ControlNet,將0號ControlNet的preprocessor和model設定為Depth,Weights設定為0.5。 - -![](../../images/sd-webui-depth-lib-4.webp) - -5. 再點選1號ControlNet,將model設定為OpenPose,Weights設定為0.5。 - -![](../../images/sd-webui-depth-lib-5.webp) - -6. 點選Generate生成,可看到圖片按照方才製作的手和骨架生成了。 diff --git a/content.zh-tw/extensions/sd-webui-mov2mov.md b/content.zh-tw/extensions/sd-webui-mov2mov.md deleted file mode 100644 index b74f0c2..0000000 --- a/content.zh-tw/extensions/sd-webui-mov2mov.md +++ /dev/null @@ -1,27 +0,0 @@ ---- -title: "AI繪圖轉影片 mov2mov" -weight: 7 ---- - -- 擴充功能來源:[Scholar01/sd-webui-mov2mov](https://github.com/Scholar01/sd-webui-mov2mov) - -*需要先安裝ControlNet才能使用這個擴充功能。 - -將影片逐一抽出畫格,使用ControlNet生圖,然後再自動合成新影片。可以設定輸出的畫格率,將人物單獨處理。 - - -1. 切換到mov2mov頁面,輸入提示詞, 再上傳影片。 - -![](../../images/mov2mov-1.webp) - -2. 在下面設定輸出影片的長寬。生圖的長寬比例應與原始影片一致。 - -![](../../images/mov2mov-2.webp) - -3. 最下面是單獨用ControlNet處理人物的選項,可以改善動作偵測的精度。 - -![](../../images/mov2mov-3.webp) - -4. 點選右上角Generate,即會開始生成。成品位於主程式下的`/outputs/mov2mov-images`目錄。 - -![](../../images/mov2mov-4.webp) diff --git a/content.zh-tw/extensions/stable-diffusion-webui-civitai-helper.md b/content.zh-tw/extensions/stable-diffusion-webui-civitai-helper.md deleted file mode 100644 index 8e65441..0000000 --- a/content.zh-tw/extensions/stable-diffusion-webui-civitai-helper.md +++ /dev/null @@ -1,54 +0,0 @@ ---- -title: "自動從CivitAI下載模型和縮圖 Civitai Extension" -weight: 5 ---- - -- 擴充功能來源:[butaixianran/Stable-Diffusion-Webui-Civitai-Helper](https://github.com/butaixianran/Stable-Diffusion-Webui-Civitai-Helper) - -喜歡從[CivitAI](https://civitai.com/)下載模型的用戶,裝這個擴充功能就不用[手動下載模型](../installation/download-models)了!此擴充功能於Stable Diffusion WebUI的界面加入CivitAI模型下載頁面,方便下載各種模型,還會自動比對資料抓取模型的縮圖。 - -安裝方法:進入Extensions頁面,點選Install from URL,輸入`https://github.com/civitai/sd_civitai_extension.git`後按Install,關掉終端機,重啟WebUI。 - -重開WebUI頁面即會多出"Civitai Helper"。 - -![](../../images/stable-diffusion-webui-civitai-helper-1.webp) - - -此頁面有4個功能: - - -## 1. 自動下載縮圖 - -點選橘色的Scan按鈕,它就會自動比對本機模型(存檔點、嵌入、超網路、LoRA)的資料,從CivitAI下載模型縮圖。 - -![](../../images/stable-diffusion-webui-civitai-helper-2.webp) - - -## 2. 從CivitAI取得模型資訊 - -此功能用於爬取本機模型的資訊。 - -於`Civitai URL`的欄位填入模型的網址(就是你用瀏覽器所看到的網址),例如這裡我選擇Deliberate這個本機模型。 - -點選`Get Model Info from Civitai`即會下載模型的資料。 - -![](../../images/stable-diffusion-webui-civitai-helper-3.webp) - - -## 3. 從CivitAI下載模型檔 - -1. 於`Civitai URL`欄位填入模型的網址(就是你用瀏覽器所看到的網址),點選`1. Get Model Info by Civitai Url`取得該頁面的資料。 - -![](../../images/stable-diffusion-webui-civitai-helper-4.webp) - -2. 它會自動抓取模型種類和資訊,`Model Name`模型名稱和`Model Version`模型版本可以自選。點選`3. Download Model`開始下載模型。至於下載進度條請看終端機。 - -![](../../images/stable-diffusion-webui-civitai-helper-5.webp) - - -## 4. 檢查模型有無更新 - -將Model Types全部打勾,點選`Check New Version from Civitai`,若有新版模型它就會自動下載。 - -![](../../images/stable-diffusion-webui-civitai-helper-6.webp) - diff --git a/content.zh-tw/extensions/stable-diffusion-webui-images-browser.md b/content.zh-tw/extensions/stable-diffusion-webui-images-browser.md deleted file mode 100644 index 21948cf..0000000 --- a/content.zh-tw/extensions/stable-diffusion-webui-images-browser.md +++ /dev/null @@ -1,18 +0,0 @@ ---- -title: "圖片庫瀏覽器 Images Browser" -weight: 5 ---- - -- 擴充功能來源:[yfszzx/stable-diffusion-webui-images-browser](https://github.com/yfszzx/stable-diffusion-webui-images-browser) - -安裝後即可方便瀏覽生圖的結果,並查看使用過的提示詞歷史紀錄。 - -安裝方法:進入Extensions頁面,點選Install from URL,輸入`https://github.com/yfszzx/stable-diffusion-webui-images-browser.git`後按Install,重啟WebUI。WebUI頁面即會多出"Images Browser"的頁面。 - -點進去,按一下First Page即會顯示最近生圖的結果。 - -![](../../images/stable-diffusion-webui-images-browser-1.webp) - -點選單張圖片,查看使用過的提示詞,還可以給予星級評分。 - -![](../../images/stable-diffusion-webui-images-browser-2.webp) diff --git a/content.zh-tw/extensions/stable-gimpfusion.md b/content.zh-tw/extensions/stable-gimpfusion.md deleted file mode 100644 index 57e7896..0000000 --- a/content.zh-tw/extensions/stable-gimpfusion.md +++ /dev/null @@ -1,30 +0,0 @@ ---- -title: "修圖軟體GIMP整合 Stable-Gimpfusion" -weight: 9 ---- - -- 擴充功能來源:[ArtBIT/stable-gimpfusion](https://github.com/ArtBIT/stable-gimpfusion) - -ArtBIT開發的Stable-Gimpfusion能讓你在GIMP修圖時的圖層調用Stable Diffusion填充內容,支援文生圖、圖生圖、ControlNet。 - -類似的擴充功能有[blueturtleai/gimp-stable-diffusion](https://github.com/blueturtleai/gimp-stable-diffusion),支援不規則選取區,還可以調用志願者經營的stable-horde伺服器來繪圖。但是它不支援SD WebUI。 - -1. 編輯SD WebUI的啟動引數,加入`--api`引數,再啟動SD WebUI。網頁界面可以不用開啟。 - -2. 安裝自由軟體[GIMP](https://www.gimp.org/) - -3. 下載[stable_gimpfusion.py](https://raw.githubusercontent.com/ArtBIT/stable-gimpfusion/main/stable_gimpfusion.py),放到GIMP的擴充功能資料夾 (Linux為`$HOME/.gimp-2.10/plug-ins/`並賦予執行權限,Windows為`C:\Users\{your_id}\AppData\Roaming\GIMP\2.10\plug-ins\`) - -4. 開啟GIMP,開啟圖片。 - -5. 點選右上角的Gimpfusion,點選要使用文生圖(text to image)還是圖生圖(image to image) - -![](../../images/stable-gimpfusion-1.webp) - -6. 輸入提示詞,Stable Diffusion就會開始生圖 - -![](../../images/stable-gimpfusion-2.webp) - -7. 進度條可從終端機查看,之後結果會新增至GIMP的圖層。目前此擴充功能只支援填充整個圖層。 - -![](../../images/stable-gimpfusion-3.webp) diff --git a/content.zh-tw/extensions/text-generation-webui.md b/content.zh-tw/extensions/text-generation-webui.md deleted file mode 100644 index e010b03..0000000 --- a/content.zh-tw/extensions/text-generation-webui.md +++ /dev/null @@ -1,8 +0,0 @@ ---- -title: "大型語言模型整合 Text Generation WebUI" -weight: 11 ---- - -與AI對話,並利用Stable Diffusion生圖。 - -參考 [Text Generation WebUI](https://ivonblog.com/posts/text-generation-webui-extensions/)的「sd_api_pictures」擴充功能。 diff --git a/content.zh-tw/installation/command-line-arguments-and-settings.md b/content.zh-tw/installation/command-line-arguments-and-settings.md index 46ff7b2..c7b5b27 100644 --- a/content.zh-tw/installation/command-line-arguments-and-settings.md +++ b/content.zh-tw/installation/command-line-arguments-and-settings.md @@ -3,9 +3,7 @@ title: "2.3. 命令列引數" weight: 2 --- -{{< hint danger >}} -此頁面僅供參考:左邊各個系統(Linux/Windows/macOS)的安裝教學已包含懶人引數,照抄即可。 -{{< /hint >}} +註:這篇文章僅供參考,本系列教學已經附上預先設定好的指令引數了,照抄即可。 Stable Diffusion WebUI的專案資料夾附有啟動主程式的指令稿(script),稱為啟動指令稿。 @@ -30,6 +28,8 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram 可在啟動指令稿使用這些環境變數(environment variables): +
+ |名稱|說明| |---|---| |PYTHON|設定Python執行檔路徑。| @@ -45,6 +45,7 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram |SD_WEBUI_CACHE_FILE|快取檔案路徑,預設值為根目錄下的`cache.json`。| |SD_WEBUI_RESTAR|由啟動指令稿(`webui.sh`或`webui.bat`)設定的值,告訴WebUI可以使用重啟功能。| |SD_WEBUI_RESTARTING|檢測WebUI是否正在重啟或重新載入的內部數值,可以用來關閉自動開啟瀏覽器的功能。設為`1`即不要自動開啟瀏覽器。設為`0`的話,即使WebUI正在重啟也自動開啟瀏覽器。| +
# 3. 命令列引數 @@ -53,65 +54,66 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram 注意下面是二條橫線「- -」 +
-| 引數指令 | 數值 | 預設值 | 說明 | +| 引數指令|數值|預設值|說明 | |---|---|---|---| | **設定值** | -|-h, --help | None | False | 顯示此幫助訊息並退出。 | -|--exit | | | 安裝後終止程式。 | -|--data-dir | DATA_DIR | ./ | 使用者資料儲存的路徑。 | -|--config | CONFIG | configs/stable-diffusion/v1-inference.yaml | 建構模型設定檔的路徑。 | -|--ckpt | CKPT | model.ckpt | Stable Diffusion模型的存檔點模型路徑。一旦指定,該模型會加入至存檔點模型列表並載入。| -|--ckpt-dir | CKPT_DIR | None | 存放Stable Diffusion模型存檔點模型的路徑。 | -|--no-download-sd-model | None | False | 即使找不到模型,也不自動下載SD1.5模型。 | -|--vae-dir | VAE_PATH | None | VAE的路徑。 | -|--gfpgan-dir| GFPGAN_DIR | GFPGAN/ | GFPGAN路徑 | -|--gfpgan-model| GFPGAN_MODEL | GFPGAN模型檔名 | -|--codeformer-models-path | CODEFORMER_MODELS_PATH | models/Codeformer/ | Codeformer模型檔的路徑。 | -|--gfpgan-models-path | GFPGAN_MODELS_PATH | models/GFPGAN | GFPGAN模型檔的路徑。 | -|--esrgan-models-path | ESRGAN_MODELS_PATH | models/ESRGAN |ESRGAN模型檔的路徑。| -|--bsrgan-models-path | BSRGAN_MODELS_PATH | models/BSRGAN |BSRGAN模型檔的路徑。 | -|--realesrgan-models-path | REALESRGAN_MODELS_PATH | models/RealESRGAN | RealESRGAN模型檔的路徑。 | -|--scunet-models-path | SCUNET_MODELS_PATH | models/ScuNET | ScuNET模型檔的路徑。 | -|--swinir-models-path | SWINIR_MODELS_PATH | models/SwinIR | SwinIR和SwinIR v2模型檔的路徑。 | -|--ldsr-models-path | LDSR_MODELS_PATH | models/LDSR | 含有LDSR模型檔的路徑。 | -|--lora-dir | LORA_DIR | models/Lora | 含有LoRA模型檔的路徑。| -|--clip-models-path | CLIP_MODELS_PATH | None | 含有CLIP模型檔的路徑。 | -|--embeddings-dir | EMBEDDINGS_DIR | embeddings/ | Textual inversion的embeddings路徑 (預設: embeddings) | -|--textual-inversion-templates-dir | TEXTUAL_INVERSION_TEMPLATES_DIR | textual_inversion_templates | Textual inversion範本的路徑| -|--hypernetwork-dir | HYPERNETWORK_DIR | models/hypernetworks/ | Hypernetwork路徑 | -|--localizations-dir | LOCALIZATIONS_DIR | localizations/ | 在地化翻譯路徑 | -|--styles-file | STYLES_FILE | styles.csv | 風格檔名 | -|--ui-config-file | UI_CONFIG_FILE | ui-config.json | UI設定檔檔名 | -|--no-progressbar-hiding | None | False | 取消隱藏Gradio UI的進度條 (我們之所以將其隱藏,是因為在瀏覽器啟動硬體加速的狀況下,進度條會降低機器學習的效能) | -|--max-batch-count| MAX_BATCH_COUNT | 16 | UI的最大批次數值 | -|--ui-settings-file | UI_SETTINGS_FILE | config.json | UI設定值畫面的檔名 | -|--allow-code | None | False | 允許在WebUI執行自訂指令稿 | -|--share | None | False | 使用此引數在啟動後會產生Gradio網址,使WebUI能從外部網路存取 | -|--listen | None | False | 以0.0.0.0主機名稱啟動Gradio,使其能回應連線請求 | -|--port | PORT | 7860 | 以給定的通訊埠啟動Gradio。1024以下的通訊埠需要root權限。如果可用的話,預設使用7860通訊埠。 | -|--hide-ui-dir-config | None | False | 在WebUI隱藏設定檔目錄。| -|--freeze-settings | None | False | 停用編輯設定。 | -|--enable-insecure-extension-access | None | False | 無視其他選項,強制啟用擴充功能頁籤。 | -|--gradio-debug | None | False | 使用 --debug選項啟動Gradio | -|--gradio-auth | GRADIO_AUTH | None | 設定Gardio授權,例如"username:password",或是逗號分隔值形式"u1:p1,u2:p2,u3:p3" | -|--gradio-auth-path | GRADIO_AUTH_PATH | None | 設定Gardio授權檔案路徑。 例如 "/路徑/" 再加上`--gradio-auth`的格式。 | -|--disable-console-progressbars | None | False | 不在終端機顯示進度條。 | -|--enable-console-prompts | None | False | 在使用文生圖和圖生圖的時候,於終端機印出提示詞 | -|--api | None | False | 以API模式啟動WebUI | -|--api-auth | API_AUTH | None | 設定API授權,例如"username:password",或是逗號分隔值形式"u1:p1,u2:p2,u3:p3" | -|--api-log | None | False | 啟用所有API請求的紀錄檔 | -|--nowebui | None | False | 僅啟動API, 不啟動WebUI | -|--ui-debug-mode | None | False | 不載入模型,以更快啟動WebUI | -|--device-id | DEVICE_ID | None | 選擇要使用的CUDA裝置 (例如在啟動指令稿使用export CUDA_VISIBLE_DEVICES=0或1) | -|--administrator | None | False | 使用系統管理員權限 | -|--cors-allow-origins | CORS_ALLOW_ORIGINS | None | 允許跨來源資源共用,列表以逗號分隔,不可有空格 | -|--cors-allow-origins-regex | CORS_ALLOW_ORIGINS_REGEX | None | 允許跨來源資源共用,後面加上單一正規表達式 | -|--tls-keyfile | TLS_KEYFILE | None | 部份啟用TLS,,需要配合--tls-certfile才能正常運作 | -|--tls-certfile | TLS_CERTFILE | None | 部份啟用TLS,需要配合--tls-keyfile才能正常運作 | -|--server-name | SERVER_NAME | None | 設定伺服器主機名稱 | -|--skip-version-check | None | False | 不檢查torch和xformers的版本 | -|--no-hashing | None | False | 停用計算存檔點模型的sha256雜湊值,加快載入速度 | +|-h, --help |None |False | 顯示此幫助訊息並退出。 | +|--exit|||安裝後終止程式。 | +|--data-dir|DATA_DIR|./|使用者資料儲存的路徑。 | +|--config |CONFIG|configs/stable-diffusion/v1-inference.yaml |建構模型設定檔的路徑。 | +|--ckpt | CKPT |model.ckpt |Stable Diffusion模型的存檔點模型路徑。一旦指定,該模型會加入至存檔點模型列表並載入。| +|--ckpt-dir | CKPT_DIR|None |存放Stable Diffusion模型存檔點模型的路徑。 | +|--no-download-sd-model|None|False|即使找不到模型,也不自動下載SD1.5模型。 | +|--vae-dir|VAE_PATH|None | VAE的路徑。 | +|--gfpgan-dir| GFPGAN_DIR|GFPGAN/ |GFPGAN路徑 | +|--gfpgan-model| GFPGAN_MODEL |GFPGAN模型檔名 | +|--codeformer-models-path|CODEFORMER_MODELS_PATH|models/Codeformer/|Codeformer模型檔的路徑。 | +|--gfpgan-models-path|GFPGAN_MODELS_PATH|models/GFPGAN|GFPGAN模型檔的路徑。 | +|--esrgan-models-path|ESRGAN_MODELS_PATH|models/ESRGAN |ESRGAN模型檔的路徑。| +|--bsrgan-models-path|BSRGAN_MODELS_PATH|models/BSRGAN |BSRGAN模型檔的路徑。 | +|--realesrgan-models-path|REALESRGAN_MODELS_PATH|models/RealESRGAN|RealESRGAN模型檔的路徑。 | +|--scunet-models-path|SCUNET_MODELS_PATH|models/ScuNET| ScuNET模型檔的路徑。 | +|--swinir-models-path|SWINIR_MODELS_PATH|models/SwinIR| SwinIR和SwinIR v2模型檔的路徑。 | +|--ldsr-models-path|LDSR_MODELS_PATH|models/LDSR | 含有LDSR模型檔的路徑。 | +|--lora-dir|LORA_DIR|models/Lora|含有LoRA模型檔的路徑。| +|--clip-models-path|CLIP_MODELS_PATH|None|含有CLIP模型檔的路徑。 | +|--embeddings-dir|EMBEDDINGS_DIR|embeddings/ |Textual inversion的embeddings路徑 (預設: embeddings) | +|--textual-inversion-templates-dir|TEXTUAL_INVERSION_TEMPLATES_DIR|textual_inversion_templates|Textual inversion範本的路徑| +|--hypernetwork-dir|HYPERNETWORK_DIR|models/hypernetworks/ |Hypernetwork路徑 | +|--localizations-dir|LOCALIZATIONS_DIR|localizations/|在地化翻譯路徑 | +|--styles-file|STYLES_FILE|styles.csv | 風格檔名 | +|--ui-config-file|UI_CONFIG_FILE| ui-config.json | UI設定檔檔名 | +|--no-progressbar-hiding|None|False |取消隱藏Gradio UI的進度條 (我們之所以將其隱藏,是因為在瀏覽器啟動硬體加速的狀況下,進度條會降低機器學習的效能) | +|--max-batch-count| MAX_BATCH_COUNT|16 |UI的最大批次數值 | +|--ui-settings-file|UI_SETTINGS_FILE|config.json | UI設定值畫面的檔名 | +|--allow-code|None|False |允許在WebUI執行自訂指令稿 | +|--share|None|False |使用此引數在啟動後會產生Gradio網址,使WebUI能從外部網路存取 | +|--listen |None|False | 以0.0.0.0主機名稱啟動Gradio,使其能回應連線請求 | +|--port|PORT|7860 | 以給定的通訊埠啟動Gradio。1024以下的通訊埠需要root權限。如果可用的話,預設使用7860通訊埠。 | +|--hide-ui-dir-config|None|False | 在WebUI隱藏設定檔目錄。| +|--freeze-settings|None|False|停用編輯設定。 | +|--enable-insecure-extension-access|None|False|無視其他選項,強制啟用擴充功能頁籤。 | +|--gradio-debug|None|False | 使用 --debug選項啟動Gradio | +|--gradio-auth|GRADIO_AUTH|None | 設定Gardio授權,例如"username:password",或是逗號分隔值形式"u1:p1,u2:p2,u3:p3" | +|--gradio-auth-path|GRADIO_AUTH_PATH|None| 設定Gardio授權檔案路徑。 例如 "/路徑/" 再加上`--gradio-auth`的格式。 | +|--disable-console-progressbars|None|False | 不在終端機顯示進度條。 | +|--enable-console-prompts|None|False | 在使用文生圖和圖生圖的時候,於終端機印出提示詞 | +|--api|None|False|以API模式啟動WebUI | +|--api-auth|API_AUTH|None|設定API授權,例如"username:password",或是逗號分隔值形式"u1:p1,u2:p2,u3:p3" | +|--api-log|None|False|啟用所有API請求的紀錄檔 | +|--nowebui|None|False|僅啟動API, 不啟動WebUI | +|--ui-debug-mode|None|False|不載入模型,以更快啟動WebUI | +|--device-id|DEVICE_ID|None|選擇要使用的CUDA裝置 (例如在啟動指令稿使用export CUDA_VISIBLE_DEVICES=0或1) | +|--administrator|None|False|使用系統管理員權限 | +|--cors-allow-origins|CORS_ALLOW_ORIGINS|None|允許跨來源資源共用,列表以逗號分隔,不可有空格 | +|--cors-allow-origins-regex|CORS_ALLOW_ORIGINS_REGEX|None|允許跨來源資源共用,後面加上單一正規表達式 | +|--tls-keyfile|TLS_KEYFILE|None|部份啟用TLS,,需要配合--tls-certfile才能正常運作 | +|--tls-certfile|TLS_CERTFILE|None|部份啟用TLS,需要配合--tls-keyfile才能正常運作 | +|--server-name|SERVER_NAME|None|設定伺服器主機名稱 | +|--skip-version-check|None|False|不檢查torch和xformers的版本 | +|--no-hashing|None|False|停用計算存檔點模型的sha256雜湊值,加快載入速度 | |--skip-version-check|None|False|不檢查torch與xformers版本。| |--skip-version-check|None|False|不檢查Python版本。| |--skip-torch-cuda-test|None|False|不檢查CUDA是否正常運作。| @@ -121,51 +123,53 @@ set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram |--api-server-stop|None|False|允許透過API通訊停止/重啟/強制停止主程式。| |--timeout-keep-alive|int|30|設定uvicorn的timeout_keep_alive數值。| | **性能相關** | -|--xformers | None | False | 給cross attention layers啟用xformers | -|--reinstall-xformers | None | False | 強制重裝xformers,升級時很有用。但為避免不斷重裝,升級後將會移除。 | -|--force-enable-xformers | None | False | 強制給cross attention layers啟用xformers ***此選項無法運作的話請勿回報bug*** | -|--xformers-flash-attention | None | False | 給xformers啟用Flash Attention,提昇再現能力 (僅支援SD2.x或以此為基礎的模型)| -|--opt-split-attention | None | False | 強制啟用Doggettx的cross-attention layer最佳化。有CUDA的系統預設啟用此選項。 | -|--opt-split-attention-invokeai | None | False | 強制啟用InvokeAI的cross-attention layer最佳化。無CUDA的系統預設啟用此選項。 | -|--opt-split-attention-v1 | None | False | 啟用舊版的split attention最佳化,防止佔用全部可用的VRAM, | -|--opt-sub-quad-attention | None | False | 啟用增進記憶體效率的sub-quadratic cross-attention layer最佳化| -|--sub-quad-q-chunk-size | SUB_QUAD_Q_CHUNK_SIZE | 1024 | sub-quadratic cross-attention layer最佳化使用的序列化區塊大小| -|--sub-quad-kv-chunk-size | SUB_QUAD_KV_CHUNK_SIZE | None | sub-quadratic cross-attention layer最佳化使用的kv區塊大小| -|--sub-quad-chunk-threshold | SUB_QUAD_CHUNK_THRESHOLD | None |sub-quadratic cross-attention layer最佳化過程中,區塊化使用的VRAM閾值| -|--opt-channelslast | None | False | 啟用4d tensors使用的alternative layout,或許可以加快推理速度 **僅適用搭載Tensor核心的Nvidia顯示卡(16xx系列以上)** | -|--disable-opt-split-attention | None | False | 強制停用cross-attention layer的最佳化 | -|--disable-nan-check | None | False | 不檢查生成圖像/潛在空間是否有nan。在CI模式無使用存檔點模型的時候很有用。| -|--use-cpu | {all, sd, interrogate, gfpgan, bsrgan, esrgan, scunet, codeformer} | None | 讓部份模塊使用CPU作為PyTorch的裝置 | -|--no-half | None | False | 不將模型轉換為半精度浮點數 | -|--precision | {full,autocast} | autocast | 使用此精度評估 | -|--no-half-vae | None | False | 不將VAE模型轉換為半精度浮點數 | -|--upcast-sampling | None | False | 向上取樣。搭配 --no-half使用則無效。生成的結果與使用--no-half引數相近,效率更高,使用更少記憶體。| -|--medvram | None | False | 啟用Stable Diffusion模型最佳化,犧牲速度,換取較小的VRAM佔用。 | -|--lowvram | None | False | 啟用Stable Diffusion模型最佳化,大幅犧牲速度,換取更小的VRAM佔用。 | -|--lowram | None | False | 將Stable Diffusion存檔點模型的權重載入至VRAM,而非RAM | +|--xformers|None|False | 給cross attention layers啟用xformers | +|--reinstall-xformers|None|False | 強制重裝xformers,升級時很有用。但為避免不斷重裝,升級後將會移除。 | +|--force-enable-xformers | None|False | 強制給cross attention layers啟用xformers ***此選項無法運作的話請勿回報bug*** | +|--xformers-flash-attention|None|False|給xformers啟用Flash Attention,提昇再現能力 (僅支援SD2.x或以此為基礎的模型)| +|--opt-split-attention|None|False | 強制啟用Doggettx的cross-attention layer最佳化。有CUDA的系統預設啟用此選項。 | +|--opt-split-attention-invokeai|None|False | 強制啟用InvokeAI的cross-attention layer最佳化。無CUDA的系統預設啟用此選項。 | +|--opt-split-attention-v1|None|False | 啟用舊版的split attention最佳化,防止佔用全部可用的VRAM, | +|--opt-sub-quad-attention|None|False|啟用增進記憶體效率的sub-quadratic cross-attention layer最佳化| +|--sub-quad-q-chunk-size|SUB_QUAD_Q_CHUNK_SIZE|1024|sub-quadratic cross-attention layer最佳化使用的序列化區塊大小| +|--sub-quad-kv-chunk-size|SUB_QUAD_KV_CHUNK_SIZE|None|sub-quadratic cross-attention layer最佳化使用的kv區塊大小| +|--sub-quad-chunk-threshold|SUB_QUAD_CHUNK_THRESHOLD|None |sub-quadratic cross-attention layer最佳化過程中,區塊化使用的VRAM閾值| +|--opt-channelslast|None|False | 啟用4d tensors使用的alternative layout,或許可以加快推理速度 **僅適用搭載Tensor核心的Nvidia顯示卡(16xx系列以上)** | +|--disable-opt-split-attention|None|False | 強制停用cross-attention layer的最佳化 | +|--disable-nan-check|None|False|不檢查生成圖像/潛在空間是否有nan。在CI模式無使用存檔點模型的時候很有用。| +|--use-cpu|{all, sd, interrogate, gfpgan, bsrgan, esrgan, scunet, codeformer}|None|讓部份模塊使用CPU作為PyTorch的裝置 | +|--no-half |None|False | 不將模型轉換為半精度浮點數 | +|--precision|{full,autocast}|autocast |使用此精度評估 | +|--no-half-vae|None|False |不將VAE模型轉換為半精度浮點數 | +|--upcast-sampling|None|False|向上取樣。搭配 --no-half使用則無效。生成的結果與使用--no-half引數相近,效率更高,使用更少記憶體。| +|--medvram |None|False |啟用Stable Diffusion模型最佳化,犧牲速度,換取較小的VRAM佔用。 | +|--lowvram |None|False |啟用Stable Diffusion模型最佳化,大幅犧牲速度,換取更小的VRAM佔用。 | +|--lowram |None|False |將Stable Diffusion存檔點模型的權重載入至VRAM,而非RAM | |--disable-model-loading-ram-optimization|None|False|停用模型載入時降低RAM佔用的優化。| | **功能** | -|--autolaunch | None | False | 啟動WebUI後自動開啟系統預設的瀏覽器 | -|--theme | None | Unset | 使用指定主題啟動WebUI (light或dark),無指定則使用瀏覽器預設主題。 | -|--use-textbox-seed | None | False | 在WebUI的種子欄位使用textbox (沒有上下,但可以輸入長的種子碼) | -|--disable-safe-unpickle | None | False | 不檢查PyTorch模型是否有惡意程式碼 | -|--ngrok | NGROK | None | Ngrok授權權杖, --share引數的替代品。 | -|--ngrok-region | NGROK_REGION | us | 選擇啟動Ngrok的區域 | +|--autolaunch|None|False | 啟動WebUI後自動開啟系統預設的瀏覽器 | +|--theme|None|Unset | 使用指定主題啟動WebUI (light或dark),無指定則使用瀏覽器預設主題。 | +|--use-textbox-seed|None|False | 在WebUI的種子欄位使用textbox (沒有上下,但可以輸入長的種子碼) | +|--disable-safe-unpickle|None|False | 不檢查PyTorch模型是否有惡意程式碼 | +|--ngrok|NGROK|None |Ngrok授權權杖, --share引數的替代品。 | +|--ngrok-region|NGROK_REGION|us |選擇啟動Ngrok的區域 | |--update-check|None|None|啟動時檢查有無新版本。| |--update-all-extensions|None|None|在啟動WebUI的時候自動更新所有擴充功能。| |--reinstall-xformers|None|False|強制重新安裝xformers,適用於更新程式之後執行。更新完之後記得移除此引數。| |--reinstall-torch|None|False|強制重新安裝touch,適用於更新程式之後執行。更新完之後記得移除此引數。| |--tests|TESTS|False|執行功能測試,確認WebUI正常運作。| -|--no-tests| None | False | 即使有`--test`引數也不要執行功能測試。 | -|--dump-sysinfo | None | False | 傾印系統訊息並退出程式(不包括擴充功能) | +|--no-tests| None|False|即使有`--test`引數也不要執行功能測試。 | +|--dump-sysinfo | None|False|傾印系統訊息並退出程式(不包括擴充功能) | |--disable-all-extensions|None|False|停用所有擴充功能,包含內建的擴充功能。| |--disable-extra-extensions|None|False|停用所有擴充功能。| | **已經無效的選項** | -|--show-negative-prompt | None | False | 無作用 | -|--deepdanbooru | None | False |無作用 | -|--unload-gfpgan | None | False | 無作用 | -|--gradio-img2img-tool | GRADIO_IMG2IMG_TOOL | None | 無作用 | -|--gradio-inpaint-tool | GRADIO_INPAINT_TOOL | None | 無作用 | -|--gradio-queue | None | False | 無作用 | -|--add-stop-route | None | False | 無作用 | -|--always-batch-cond-uncond | None | False | 無作用 | +|--show-negative-prompt|None|False | 無作用 | +|--deepdanbooru|None|False |無作用 | +|--unload-gfpgan|None|False |無作用 | +|--gradio-img2img-tool|GRADIO_IMG2IMG_TOOL|None|無作用 | +|--gradio-inpaint-tool|GRADIO_INPAINT_TOOL|None|無作用 | +|--gradio-queue|None|False|無作用 | +|--add-stop-route|None|False|無作用 | +|--always-batch-cond-uncond|None|False|無作用 | + +
diff --git a/content.zh-tw/installation/deploy-to-paperspace.md b/content.zh-tw/installation/deploy-to-paperspace.md deleted file mode 100755 index 93624b6..0000000 --- a/content.zh-tw/installation/deploy-to-paperspace.md +++ /dev/null @@ -1,47 +0,0 @@ ---- -title: "☁️ 部署至Paperspace" -weight: 12 ---- - -Paperspace是一間位於美國的雲端機器學習服務供應商,提供免費GPU跑機器學習專案。 - -免費版本的儲存空間5GB,只能使用有限制的執行個體,一次只能執行一個筆記本,6小時後機器自動關閉。有時候免費GPU需要排隊。 - -值得一提的是Paperspace官方有維護專門跑Stable Diffusion WebUI的筆記本,點一下就可以使用,並且速度很快。程式碼很清晰,可按照需求修改。 - - -1. 至[Paperspace](https://www.paperspace.com/)註冊帳號。註冊需要提供電話號碼。 - -2. 進入[gradient-ai/stable-diffusion](https://console.paperspace.com/github/gradient-ai/stable-diffusion?machine=Free-GPU&ref=blog.paperspace.com)頁面,點選Start Machine - -![](../../images/deploy-to-paperspace-1.webp) - -3. 這個面板就是筆記本的頁面了。點選右上角的Run all後,筆記本會安裝Stable Diffusion WebUI所需的依賴,並下載Stable Diffusion v2.1的模型。此階段所下載的檔案不會計入帳戶的儲存空間。 - -![](../../images/deploy-to-paperspace-2.webp) - - -4. 之後點選Use the Stable Diffusion Web UI這個儲存格終端機顯示的Gradio網址 - -![](../../images/deploy-to-paperspace-3.webp) - - -5. 即可使用WebUI。 - -![](../../images/deploy-to-paperspace-4.webp) - -6. 使用完畢後記得按左上角的Stop Machine關機。 - - -此筆記本只有安裝最小功能。擴充功能和額外模型要自行修正筆記本程式下載。 - -下載自訂模型可以加在複製Stable Diffusion WebUI的儲存庫之後 -```python -!git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui -!wget -nc -P stable-diffusion-webui/models/Stable-diffusion https://huggingface.co/andite/anything-v4.0/resolve/main/anything-v4.5-pruned.safetensors -``` - -參照[命令列引數](../installation/command-line-arguments-and-settings/),找到啟動WebUI這行,加入`--enable-insecure-extension-access`讓你可從網頁版的Extension頁面下載擴充。 -```python -!python launch.py --share --config ~/../datasets/stable-diffusion-classic-v2/768-v-ema.yaml --ckpt ~/../datasets/stable-diffusion-classic-v2/768-v-ema.ckpt --enable-insecure-extension-access -``` diff --git a/content.zh-tw/installation/download-models.md b/content.zh-tw/installation/download-models.md index fb51df9..1eb1231 100644 --- a/content.zh-tw/installation/download-models.md +++ b/content.zh-tw/installation/download-models.md @@ -20,9 +20,8 @@ Stable Diffusion WebUI在初次啟動時,若放模型的資料夾(`models/Stab 另一個是[Civitai](https://civitai.com/),專門用來分享Stable Diffusion相關的資源,特色是模型都有示範縮圖,用戶也可以分享彼此使用的提示詞,以及分享作品。 -{{< hint danger >}} -Civitai可能包含未成年不宜的圖像。 -{{< /hint >}} +Civitai可能包含未成年不宜的圖像。 + ## 寫實風模型 @@ -38,21 +37,19 @@ Civitai可能包含未成年不宜的圖像。 [Deliberate](https://civitai.com/models/4823/deliberate):基於SD-1.5模型,適合生成精緻寫實風格的人物、動物、自然風景。 -[Realistic Vision v1.4](https://civitai.com/models/4201/realistic-vision-v13):寫實風人物與動物模型。 +[Realistic Vision v6.0](https://civitai.com/models/4201/realistic-vision-v60-b1):寫實風人物與動物模型。 ## 動漫風模型 -[Anything萬象熔爐 v4.5](https://huggingface.co/andite/anything-v4.0)適合畫動漫圖,作者宣稱不需要打一堆提示詞也能出漂亮的圖。 - -[Waifu Diffusion v1.4](https://huggingface.co/hakurei/waifu-diffusion-v1-4)是純粹使用Danbooru圖庫訓練而成,適合畫動漫圖。 - -[Hentai Diffusion](https://github.com/Delcos/Hentai-Diffusion)適合畫動漫圖,模型已使用大量負向提示詞訓練過以排除不良結果,另提供embeddings方便繪圖時使用。 +[Anything萬象熔爐 v5](https://civitai.com/models/9409/anything-or)適合畫動漫圖,作者宣稱不需要打一堆提示詞也能出漂亮的圖。 [DreamShaper](https://civitai.com/models/4384/dreamshaper)是基於SD-1.5模型,生成精細動漫人物與油畫風格的模型。 [OrangeMix3](https://huggingface.co/WarriorMama777/OrangeMixs),混合多種風格的動漫繪圖模型,偏寫實。 +[Waifu Diffusion v1.4](https://huggingface.co/hakurei/waifu-diffusion-v1-4)是純粹使用Danbooru圖庫訓練而成,適合畫動漫圖。 + # 2. 如何下載模型 @@ -64,7 +61,6 @@ HuggingFace為儲存庫 → Files and versions,選擇下載副檔名為`.ckpt` Civitai則是點選Download按鈕下載模型。 - 如果要用指令下載,可以用wget或git,例如下載`Anything V4.5`的模型到`stable-diffusion-webui/models/Stable-diffusion`資料夾: ```bash wget -nc -P stable-diffusion-webui/models/Stable-diffusion https://huggingface.co/andite/anything-v4.0/resolve/main/anything-v4.5-pruned.safetensors diff --git a/content.zh-tw/installation/system-requirements.md b/content.zh-tw/installation/system-requirements.md index 8a86289..6c4f626 100644 --- a/content.zh-tw/installation/system-requirements.md +++ b/content.zh-tw/installation/system-requirements.md @@ -53,15 +53,13 @@ weight: 1 - 要生成513x512的圖,顯示卡VRAM在4GB以下的很容易遇到記憶體不足的問題,並且生圖速度很慢。 -- 儘管可以[用純CPU跑Stable Diffusion WebUI](https://ivonblog.com/posts/stable-diffusion-running-on-cpu/),但是速度會非常慢。一張顯示卡30秒就能算好的圖,CPU要算10分鐘。因此用獨立顯示卡跑Stable Diffsuion仍是比較好的選擇。 +- 儘管可以[用純CPU跑Stable Diffusion WebUI](https://ivonblog.com/posts/stable-diffusion-running-on-cpu/),但是速度會非常慢。一張顯示卡30秒就能算好的圖,CPU要算10分鐘。因此用獨立顯示卡跑Stable Diffusion仍是比較好的選擇。 - 硬體不夠力的話,用雲端服務跑也是可以,例如Google Colab。 # 2. 作業系統需求 -## GNU/Linux - 測試可以跑的Linux發行版: - Debian 12 diff --git a/content.zh-tw/tutorials/text-to-image-controlnet-latent-couple.md b/content.zh-tw/tutorials/text-to-image-controlnet-latent-couple.md index 7f59f1e..09de648 100644 --- a/content.zh-tw/tutorials/text-to-image-controlnet-latent-couple.md +++ b/content.zh-tw/tutorials/text-to-image-controlnet-latent-couple.md @@ -3,7 +3,7 @@ title: "用文生圖 + ControlNet + Latent Couple生成指定位置的多個 weight: 1 --- -請先閱讀[文生圖用法](../features/text-to-image/),並安裝[ControlNet](../extensions/controlnet/)、[Latent Couple](../extensions/latent-couple/)、[PoseX](../extensions/posex/)擴充功能。 +請先閱讀[文生圖用法](../features/text-to-image/),並安裝ControlNet、Latent Couple、PoseX擴充功能。 ![](../../images/text-to-image-controlnet-latent-couple-7.webp)