diff --git "a/markdown/unit3/01_stable_diffusion_introduction_CN.md" "b/markdown/unit3/01_stable_diffusion_introduction_CN.md" --- "a/markdown/unit3/01_stable_diffusion_introduction_CN.md" +++ "b/markdown/unit3/01_stable_diffusion_introduction_CN.md" @@ -1,15 +1,15 @@ # 简介 -本节笔记本将会涵盖一些基础性的内容，介绍如何利用现有的管线（pipeline）借助 Stable Diffusion 模型来创造和修改图片。我们也将简要一览管线内的关键组成部分，把更进一步的探究任务留给更深入介绍的一节笔记本中。特别地，我们将涵盖以下内容： -- 利用 `StableDiffusionPipeline` 根据文字描述生成图片，并通过修改各个输入参数来进行探究实验 -- 实操了解管线的一些关键组成部分： +本节将会涵盖一些基础内容，介绍如何利用现有的管线（pipeline）借助 Stable Diffusion 模型来生成和修改图片。我们也将简要介绍管线内的关键组成部分。总的来讲，本节涵盖以下内容： +- 利用 `StableDiffusionPipeline` 根据文字描述生成图片，并探究各个输入参数的意义 +- 用代码来了解管线的一些关键组成部分： - 让这个模型成为“隐编码扩散模型（latent diffusion model）”的可变分自编码器（VAE） - 处理文本提示的分词器（tokenizer）和文本编码器 - - UNet 模型本身 + - UNet 模型 - 使用的调度器（scheduler），以及其它不同的调度器 - 使用管线的组成部分来复现采样循环 -- 用Img2Im管线来编辑现有图片 -- 使用inpainting管线和Depth2Img管线 +- 用 Img2Img 管线来编辑现有图片 +- 使用 inpainting 管线和 Depth2Img 管线 ❓如果你有问题，请在 Hugging Face Discord 的 `#diffusion-models-class` 频道提出。如果你还没有 Hugging Face 的账号，你可以在这里注册：https://huggingface.co/join/discord @@ -70,7 +70,7 @@ device = ( # 从文本生成图像 -我们先载入 Stable Diffusion 的管线，看看我们能做点什么。现有的 Stable Diffusion 模型有好多不同版本，截至本文书写时的最新版本是第2.1版。如果你想探究更旧的版本，只需要在 `model_id` 处修改即可（比如你可以试试将其改成 `CompVis/stable-diffusion-v1-4` 或从[dreambooth concepts library](https://huggingface.co/sd-dreambooth-library)选一个模型）。 +我们先载入 Stable Diffusion 的管线。现有的 Stable Diffusion 模型有好多不同版本，截至本书编写时的最新版本是第2.1版。修改版本只需要在 `model_id` 处修改即可（比如你可以试试将其改成 `CompVis/stable-diffusion-v1-4` 或从[dreambooth concepts library](https://huggingface.co/sd-dreambooth-library)选一个模型）。 ```python @@ -79,8 +79,9 @@ model_id = "stabilityai/stable-diffusion-2-1-base" pipe = StableDiffusionPipeline.from_pretrained(model_id).to(device) ``` -如果你的GPU内存不够用，这里有些办法也许可以减少内存使用： -- 载入 FP16 精度的版本（但并不是所有的系统上都支持）。与此同时，在你对管线的某个特定部分实验时，你也需要把所有的张量换成 torch.float16 精度： +如果你的GPU内存不够用，可以尝试这些方法减少内存使用： + +- 载入 FP16 精度的版本（但并不是所有的系统上都支持），此时你也需要保证所有的张量也被改成了 torch.float16 精度： `pipe = StableDiffusionPipeline.from_pretrained(model_id, revision="fp16", torch_dtype=torch.float16).to(device)` @@ -90,7 +91,7 @@ pipe = StableDiffusionPipeline.from_pretrained(model_id).to(device) `pipe.enable_attention_slicing()` -- 降低要生成的图片的尺寸 +- 降低生成结果的图片尺寸 当管线加载好了以后，我们可以用以下代码去使用文字提示生成图片： @@ -126,17 +127,17 @@ pipe_output.images[0] -**练习：** 花点时间去用上面的代码块实验，使用你自己的文字提示，反复调整各项设置看看这些设置如何影响生成效果。使用不同的随机种子或者移除掉 `generator` 这个输入参数，看看能获取什么不同的效果。 +**练习：** 用上面的代码块做实验，使用你自己的文字提示，反复调整各项设置看看这些设置如何影响生成效果。使用不同的随机种子或者移除掉 `generator` 这个输入参数，看看能获取什么不同的效果。 主要的要调节参数介绍： -- `width` 和 `height` 指定了生成图片的尺寸。它们必须是可被 8 整除的数字，只有这样我们的可变分自编码器（VAE）才能正常工作（我们在将来的章节会了解到）。 -- 步数 `num_inference_steps` 也会影响生成的质量。默认设成 50 已经很好了，但有些时候你也可以用少到像 20 步这样，这对做实验就方便多了。 -- 使用 `negative_prompt` 来强调不希望生成的内容，一般会在无分类器引导（classifier-free guidance）的过程中用到，这可以是个非常有用的添加额外控制的方式。你可以留空这个地方不管，但很多用户觉得列出一些不想要的特性对更好的生成很有帮助。 +- `width` 和 `height` 指定了生成图片的尺寸。它们必须是可被 8 整除的数字，��有这样我们的可变分自编码器（VAE）才能正常工作（原因后面会将）。 +- 步数 `num_inference_steps` 也会影响生成的质量。默认设成 50 就行，但你也可以试试设成 20，看看设置小了后效果会如何。 +- 使用 `negative_prompt` 来强调不希望生成的内容，一般会在无分类器引导（classifier-free guidance）的情况下用到，这种添加额外控制的方式特别管用。列出一些不想要的特性对生成更好结果很有帮助。 - `guidance_scale` 这个参数决定了无分类器引导（CFG）的影响强度有多大。增大这个值会使得生成的内容更接近文字提示；但这个值如果过大，可能会使得结果变得过饱和、不好看。 -如果你想为文字提示找点灵感，你也可以从这里开始：[Stable Diffusion Prompt Book](https://stability.ai/sdv2-prompt-book) +如果你想为文字提示找点灵感，可以参考这里：[Stable Diffusion Prompt Book](https://stability.ai/sdv2-prompt-book) -你可在下面看到增大`guidance_scale` 这个参数所带来的作用： +下面展示了增大 `guidance_scale` 这个参数所带来的作用： ```python @@ -168,8 +169,7 @@ for i, ax in enumerate(axs): ![png](01_stable_diffusion_introduction_CN_files/01_stable_diffusion_introduction_CN_12_3.png) - -调一调上面的值，尝试不同的幅度和提示。当然了，如何解读这些参数是个很主观的事情，但对我来说，我觉得 8 到 12 这个值区间比其它情况产出的结果都好。 +当然了，如何解读这些参数是个很主观的事情，但总的来说，设置值到 8 到 12 这个区间内是个不错的选择。 # 管线的组成部分 @@ -183,15 +183,15 @@ print(list(pipe.components.keys())) # List components ['vae', 'text_encoder', 'tokenizer', 'unet', 'scheduler', 'safety_checker', 'feature_extractor'] -为了更好地理解管线如何工作，我们简要地一个一个看看各个组成部分，然后我们自己动手，把它们组合在一起，以此复现出整个管线功能。 +为了更好地理解管线如何工作，我们简要地逐个了解各个组成部分。然后可以自己用代码实现，并把它们组合在一起，以此复现出整个管线功能。 ### 可变分自编码器（VAE） ![vae_diagram.png](data:image/png;base64,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) -可变分自编码器（VAE）是一种模型，它可以将输入编码成一种被压缩过的表示形式，再把这个“隐式的”表示形式解码成某种接近输入的输出。当我们使用 Stable Diffusion 生成图片时，我们先在VAE的“隐空间”应用扩散过程**生成隐编码**，然后**在结尾对它们解码**来查看结果图片。 +可变分自编码器（VAE）是一种模型，它可以将输入图像进行编码，得到一个“压缩过”的信息，再把这个“隐式的”压缩信息给解码出来，得到某种接近输入的输出。当我们使用 Stable Diffusion 生成图片时，我们先在VAE的“隐空间”应用扩散过程**生成隐编码**，然后**在结尾对它们解码**来得到最终结果图片。这样UNet的输入就不是完整的图片、而是缩小版的特征，可以极大地减少计算资源的使用。 -这里就是一个例子，使用VAE把输入图片编码成隐式的表示形式，再对它解码： +下面代码就使用了 VAE，把输入图片编码成隐式表示形式，再对它解码： ```python @@ -215,7 +215,7 @@ print("Decoded images shape:", decoded_images.shape) Decoded images shape: torch.Size([1, 3, 512, 512]) -如你所见，原本 512x512 尺寸的图片被压缩到 64x64的隐式表示形式（有四个通道）中。每一空间维度上都被压缩到了原有的八分之一，这也是为什么我们设定 `width` 和 `height` 时需要它们是 8 的倍数。 +在这个例子中，原本 512x512 尺寸的图片被压缩到 64x64的隐式表示形式（有四个通道）中。每一空间维度上都被压缩到了原有的八分之一，这也是上文中为什么我们设定 `width` 和 `height` 时需要它们是 8 的倍数。 使用这些信息量充裕的 4x64x64 隐编码可比使用 512px 大小的图片要高效多了，可以让我们的扩散模型更快，并使用更少的资源来训练和使用。VAE的解码过程并不是完美的，但即使损失了一点点质量，总的来说也足够好了。 @@ -227,9 +227,9 @@ print("Decoded images shape:", decoded_images.shape) ![text_encoder.png](data:image/png;base64,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) -文本编码器的作用是将输入的字符串（文本提示）转化成数值表示形式，这样才能输入进 UNet 作为条件。文本首先要被管线中的分词器（tokenizer）转换成一系列的分词（token）。文本编码器有大约五万分词的词汇量 —— 任何不存在于这些词汇量中的词语都会被细分成多个更小的词语。这些分词然后就被送入文本编码器模型中 —— 文本编码器是一个transformer模型，最初被训练作为CLIP的文本编码器。这里我们希望这个经过了预训练的transformer模型学习到了足够好的文本表示能力，可以对我们这里的扩散任务一样有用。 +文本编码器的作用是将输入的字符串（文本提示）转化成数值表示形式，这样才能输入进 UNet 作为条件。文本首先要被管线中的分词器（tokenizer）转换成一系列的分词（token）。文本编码器有大约五万分词的词汇量 —— 它可以将一个长句子分解为一个个的文本最小单元，这些最小单元被称为token，在英文中，一般一个单词就可以作为一个分词，而在汉语中，一般一个或多个汉字组成的词语才是一个分词。这些分词的词嵌入（embedding）会被送入文本编码器模型中 —— 文本编码器是一个transformer模型，被训练作为 CLIP 这个模型的文本编码器。在这里，这个预训练过的 transformer 模型已经有了足够好的文本表示能力，能够很好地把文本描述映射到特征空间。 -我们这里通过对一个文字提示进行编码，验证一下这个过程。首先，我们手动进行分词，并将它输入到文本编码器中，再使用管线的 `_encode_prompt` 方法，观察一下完成的过程，这包括补全或截断分词串的长度，使得分词串的长度等于最大长度 77 ： +我们这里通过对一个文本描述进行编码，验证一下这个过程。首先，我们手动进行分词，并将它输入到文本编码器中，再使用管线的 `_encode_prompt` 方法，调用一下这个编码过程，该过程包括补全或截断分词串的长度，使得分词串的长度等于最大长度 77 ： ```python @@ -274,13 +274,13 @@ text_embeddings.shape -这些文本嵌入（text embedding），也即文本编码器中最后一个transformer模块的“隐状态（hidden state）”，将会被送入 UNet 中作为 `forward` 函数的一个额外输入，下面部分我们会详细看到。 +这些文本嵌入（text embedding），也即文本编码器中最后一个transformer模块的“隐状态（hidden state）”，将会被送入 UNet 中作为 `forward` 函数的一个额外输入，稍后我们会详细看到。 ### UNet ![unet.png](data:image/png;base64,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) -UNet 模型接收一个带噪的输入，并预测噪声，和我们之前单元中看到的 UNet 一样。但与以往例子不同的是，这里的输入并不是图片了，而是图片的隐式表示形式（latent representation）。此外，除了把用于暗示带噪程度的timestep输入进 UNet 作为条件外，这里模型也把文字提示（prompt）的文本嵌入（text embeddings）作为了额外输入。这里我们假数据试着让它预测一下： +在扩散模型中，UNet 模型接收一个带噪的输入，并预测噪声，以此实现去噪。但与以往例子不同的是，这里的输入并不是图片了，而是图片的隐式表示形式（latent representation）。此外，除了把用于暗示带噪程度的 timestep 输入进 UNet 作为条件外，这里模型也把文字提示（prompt）的文本嵌入（text embeddings）也作为输入。这里我们用假数据试着让它预测一下，请注意这一过程中各个输入输出的形状大小： ```python @@ -300,9 +300,9 @@ print('UNet output shape:', unet_output.shape) # Same shape as the input latents ### 调度器（Scheduler） -调度器保存了如何加噪的计划安排，管理着如何基于模型的预测更新带噪样本。默认的调度器是 `PNDMScheduler` 调度器，但你也可以用其它的（比如 `LMSDiscreteScheduler` 调度器），只要它们用相同的配置初始化。 +调度器保存了如何加噪这一信息，管理着如何基于模型的预测更新带噪样本。默认的调度器是 `PNDMScheduler` 调度器，但你也可以用其它的（比如 `LMSDiscreteScheduler` 调度器），只要它们用相同的配置初始化。 -我们可以画出图像来观察随着timestep添加噪声的计划安排，看看不同时间的噪声水平（基于$\bar{\alpha}$这个参数）是什么样的： +我们可以画出图像来观察如何随着 timestep 添加噪声，看看噪声水平（基于$\bar{\alpha}$这个参数）是怎样变动的： ```python @@ -317,7 +317,7 @@ plt.title('Noise schedule');plt.legend(); -如果你想尝试不同的调度器，你可以像下面代码中一样换一个新的： +如果你想尝试不同的调度器，可以参考下面代码修改： ```python @@ -368,7 +368,7 @@ pipe(prompt="Palette knife painting of an winter cityscape", height=480, width=4 ### DIY一个采样循环 -现在我们已经一个个看过这些组成部分了，我们可以把它们拼装到一起，来复现一下整个管线的功能： +现在我们已经逐个了解这些组成部分了，我们可以把它们拼装到一起，来复现一下整个管线的功能： ```python @@ -424,15 +424,15 @@ pipe.numpy_to_pil(image)[0] -大多数情况下，还是使用现有的管线更方便，但我们这里手动拼装采样循环会更有益于我们理解每个部分如何工作、以及如何根据需要修改这些组件。如果你想看看实际的代码以及深入了解如何修改这些组成部分，你可以参考 Stable Diffusion Deep Dive 这个[笔记本](https://github.com/fastai/diffusion-nbs/blob/master/Stable%20Diffusion%20Deep%20Dive.ipynb)和[视频](https://m.youtube.com/watch?v=0_BBRNYInx8)。这里有对相关内容更全面彻底的探究。 +当然大多数情况下，还是使用现有的管线更方便。如果你想看看实际的代码以及深入了解如何修改这些组成部分，你可以参考 Stable Diffusion Deep Dive 这个[笔记本](https://github.com/fastai/diffusion-nbs/blob/master/Stable%20Diffusion%20Deep%20Dive.ipynb)和[视频](https://m.youtube.com/watch?v=0_BBRNYInx8)。这里有对相关内容更全面彻底的探究。 # 其它的一些管线 -那我们除了从文字提示生成图片外还能做点什么呢？其实还有很多！这一部分还会向你展示几个很酷的管线，让你了解一些其它的可以应用 Stable Diffusion 的任务。这里面有几个管线需要你下载新的模型，所以你要是着急的话你也可以跳过这一部分，只看看现有的输出展示就行，不必亲自下载和运行模型。 +除了从文字提示生成图片外，我们还能做点什么呢？其实还有很多！下面还会向你展示几个管线，让你了解一些 Stable Diffusion 的应用。 ## Img2Img -直到现在，我们生成的图片还都是完全从随机的隐变量来开始生成的，而且也都使用了完整的扩散模型采样循环。但其实我们不必从头开始。Img2Img 这个管线首先将一张已有的图片进行编码，编码成一系列的隐变量，然后在这些隐变量上随机加噪声，以这些作为起始点。噪声加多大量、去噪需要的步数决定了这个 img2img 过程的“强度”。只加一点点噪声（强度低）只会带来微小改变，而加入最大量的噪声并跑完完整的去噪过程又会生成出几乎完全不像原始图片的结果，即使可能在整体结构上还多少有点相似。 +直到现在，我们生成的图片还都是完全从随机的隐变量来开始生成的，而且也都使用了完整的扩散模型采样循环。但如果使用 Img2Img 管线，我们其实不必从头开始。Img2Img 这个管线首先将一张已有的图片进行编码，编码成一系列的隐变量，然后在这些隐变量上随机加噪声，以这些作为起始点。噪声加多大量、去噪需要多少步数（timestep）决定了这个 img2img 过程的“强度”。只加一点点噪声（强度低）只会带来微小改变，而加入大量噪声并跑完完整去噪过程又会让生成出几乎完全不像原始图片，仅在整体结构上还有点相似。 这个管线无需什么特殊模型，只要模型的 ID 和我们的文字到图像模型一样就行，没有新的需要下载的文件。 @@ -447,7 +447,7 @@ img2img_pipe = StableDiffusionImg2ImgPipeline.from_pretrained(model_id).to(devic Fetching 16 files: 0%| | 0/16 [00:00