3个插件,轻松绘制多人画面 - 使用Latent Couple、Composable LoRA和ControlNet完成多人物场景制图 – Stable Diffusion教程

小薇 Official Channel
6 Apr 202305:50

TLDR本视频介绍了如何使用Stable Diffusion绘图工具中的三个插件:Latent Couple、Composable LoRA和ControlNet,来创建多人物场景图像。Latent Couple插件通过区域分隔和权重分配来控制人物位置和数量;Composable LoRA插件允许为每个人物匹配独立的提示词,实现个性化细节;ControlNet则用于精确控制人物动作和姿势。通过这些插件,用户可以轻松绘制多人合照,自定义每个人物的服装、发型和姿势,实现高度个性化的图像创作。

Takeaways

  • 😀 使用Stable Diffusion绘图时,可以通过添加如'2girls', '3 characters'等词条来生成多人物图像,但这种方法有局限性。
  • 🔧 与midjourney等平台不同,Stable Diffusion支持多种插件,可以更精确地控制人物的位置、动作和组合方式。
  • 🎨 Latent Couple插件通过区域分隔多个人物,可以通过WebUI在线安装并调整Division、Positions和weights等参数。
  • 🖌️ Composable LoRA插件允许为每个人物匹配独立的提示词,与Latent Couple配合使用,增强多人物创作的灵活性。
  • 🖼️ 实际应用中,首先选择一个checkpoint模型,如MIX-Pro-V4,并启用相关插件,然后设置正反提示词进行图像生成。
  • 👩‍👩‍👧 通过设置不同的Division和Positions参数,可以控制人物在图像中的左右位置和顺序。
  • 👥 在三人合影效果测试中,通过调整Division为1:3和Positions参数,可以安排三个人物的位置。
  • 🎭 Controlnet插件可以精确控制人物的动作和姿势,通过上传图像和生成骨架图来应用素材图中人物的姿势。
  • 👗 使用Controlnet时,可以在全图提示词中加入'full body'词条,以显示人物全身。
  • 🌟 通过这些插件,用户可以固定人物位置,自定义服装、头发和姿势,创造出风格多样的多人物场景图像。

Q & A

  • 1. 如何通过提示词生成多人合照的图像?

    -可以在提示词中添加类似‘2girls’或‘3 characters’的词条,但这种方法在生成和用户预期一致的图像时有局限性。

  • 2. Latent Couple 插件的作用是什么?

    -Latent Couple 插件可以通过区域分隔来控制多个人物的位置和组合方式,使图像中的人物排列更精确。

  • 3. 如何安装 Latent Couple 插件?

    -进入 WebUI,搜索 stable-diffusion-webui-two-shot,在线安装后重启 UI 即可。

  • 4. Latent Couple 插件中的 Division、Positions 和 Weights 分别代表什么?

    -Division 代表画面的分隔比例,Positions 代表人物的位置,Weights 代表每个人物在画面中的权重。

  • 5. Composable LoRA 插件的功能是什么?

    -Composable LoRA 插件可以为每个人物匹配独立的提示词,特别适合与 Latent Couple 配合使用,来更细致地控制每个人物的外观和位置。

  • 6. 如何安装和使用 Composable LoRA 插件?

    -进入 WebUI,选择 'install from URL',输入安装地址并点击 'install' 即可。安装后与 Latent Couple 插件配合使用。

  • 7. 如何通过 Latent Couple 和 Composable LoRA 插件生成不同发色的两个人物?

    -在提示词中,使用 'AND' 来区分两个人物,并分别设置提示词,例如设置人物1的头发为黑色,人物2的头发为金色。

  • 8. 在生成三个人物的合影时如何设置 Division 参数?

    -将 Division 设置为 1:3,表示将屏幕分为三等分,同时调整 Positions 和 Weights 参数来控制人物的位置和权重。

  • 9. ControlNet 插件的主要功能是什么?

    -ControlNet 插件可以通过上传图像生成骨架图,精准控制人物的动作和姿势,从而生成姿势相似的图像。

  • 10. 如何通过 ControlNet 插件生成三人合影的姿势图像?

    -上传图像,选择 OpenPose 生成骨架图,分别更换人物的发型和服装,最终生成姿势与原图相似的三人合影图像。

Outlines

00:00

🎨 'Multi-Character Creation with Plugins in Stable Diffusion'

This paragraph introduces the use of Stable Diffusion for generating images of multiple characters. It highlights the limitations of using simple prompts like '2girls' or '3 characters' and suggests that while these can work on platforms without plugins, Stable Diffusion's extensive plugin support allows for more precise control over character positioning, actions, and combinations. The video presents three plugins: Latent Couple for dividing the image into regions for multiple characters, Composable LoRA for assigning independent prompts to each character, and Controlnet for controlling character poses. The paragraph explains how to install and use these plugins, including setting up division ratios, positions, weights, and visualizing the image segmentation. It also demonstrates how to apply these plugins with a checkpoint model to create images of two or three characters with specific attributes like hair color, and how to adjust parameters for different group compositions.

05:01

🖌️ 'Customizing Character Images with Controlnet and Openpose'

The second paragraph delves into the application of Controlnet and Openpose for precise control over character poses in images. It describes how to upload an image, generate a skeleton using Openpose, and then apply this skeleton to control the pose of characters in the generated image. The paragraph also discusses the ability to customize characters by changing their hairstyles and clothing while maintaining the original pose. The video concludes with a mention of providing the plugins used in the video for viewers, encourages viewers to like, subscribe, and share the content, and invites them to leave comments or contact the host with any questions. It also promotes other programs on the channel and bids farewell to the viewers.

Mindmap

Keywords

💡Stable Diffusion

Stable Diffusion is an AI model that generates images from text prompts. In the context of the video, it is the primary tool used for creating multi-character scenes. The script discusses how to enhance its capabilities with plugins to better control the composition and positioning of characters in the generated images.

💡Latent Couple

Latent Couple is a plugin for Stable Diffusion that allows for the separation of multiple characters within an image. It is used to control the division of the image into different areas, the positioning of the characters, and the weighting of each character's presence in the image. The script provides an example of how to use Latent Couple to generate a合照 of two characters by adjusting the division and positions parameters.

💡Composable LoRA

Composable LoRA is another plugin for Stable Diffusion that complements Latent Couple by enabling the association of independent prompts with each character. This allows for a more detailed customization of each character's appearance and attributes. The script illustrates how to use Composable LoRA alongside Latent Couple to create a scene with distinct characteristics for each character, such as different hair colors.

💡ControlNet

ControlNet is a plugin that provides precise control over the posture and movements of characters in the generated images. It can be used to upload a reference image, generate a skeleton, and then apply the posture to the characters in the artwork. The video script describes how ControlNet can be used to ensure that the characters maintain a consistent pose across different images.

💡WebUI

WebUI refers to the web-based user interface for Stable Diffusion, where users can interact with the AI model through a browser. The script mentions accessing WebUI to install plugins like Latent Couple and Composable LoRA, which are essential for the multi-character scene creation process discussed in the video.

💡Division

In the context of the video, 'Division' refers to the proportional division of the image space for placing multiple characters. It is a parameter within the Latent Couple plugin that determines how the image is split to accommodate each character, with examples given such as 1:1 for equal division or 1:2 for a more dominant character.

💡Positions

Positions is a parameter within the Latent Couple plugin that specifies the order and location of the characters within the divided image areas. The script provides default values and explains how to adjust these values to control the placement of characters in the final image.

💡Weights

Weights in the video script refers to the visual importance given to different elements of the image, such as characters and background, within the Latent Couple plugin. It helps to balance the prominence of each character and the background in the final composition.

💡Visualize

Visualize is a feature within the Latent Couple plugin that allows users to preview how the image will be divided based on the set parameters. It helps in understanding the spatial arrangement of characters before the image is generated.

💡Checkpoint Model

A checkpoint model in the video refers to a specific version or state of the AI model that has been saved and can be loaded for use. The script mentions selecting and enabling a checkpoint model like MIX-Pro-V4 to apply the plugins and generate multi-character images.

💡Full Body

The term 'Full Body' is used in the script to describe a prompt that ensures the generated characters are depicted in their entirety, showing the whole body. This is important for maintaining consistency in the portrayal of characters, especially when using ControlNet to match poses from reference images.

Highlights

使用Stable Diffusion绘图时,可以通过添加词条如'2girls', '3 characters'等来生成多人物图像。

Stable Diffusion的多人物生成方法存在局限性,即使输入详细提示词也难以完全符合预期。

Stable Diffusion拥有大量插件,可以更精确控制人物位置、动作和组合方式。

介绍了三种适合多人物创作的插件:Latent Couple、Composable LoRA和ControlNet。

Latent Couple插件通过区域分隔多个人物,可以控制画面分隔比例和人物位置。

Composable LoRA插件用于分隔多个人物,并为每个人物匹配独立的提示词。

ControlNet插件可以精确控制人物的动作和姿势。

Latent Couple插件的Division选项代表画面分隔比例,Positions代表人物顺序和位置。

Composable LoRA插件与Latent Couple配合使用,为每个人物添加独立的提示词。

通过设置正反提示词,可以生成具有特定特征的多人物图像。

ControlNet插件通过上传图像和生成骨架图来应用素材图中人物的姿势。

使用ControlNet插件时,可以在全图提示词中加入'full body'词条以显示人物全身。

通过更换人物的发型和服装,可以生成姿势与原图一致的图像。

使用这些插件可以固定人物位置,并自定义每个人物的服装、头发和姿势。

教程提供了如何安装和使用这些插件的详细步骤。

视频最后提供了插件的下载链接,并鼓励观众点赞、订阅和转发。