🐸This video explains how to install and use "Control-Net," an extension for the image generation AI "Stable Diffusion webUI AUTOMATIC1111" (local version). It is based on installation instructions as of March 2011.
🐸Please note that due to frequent updates by the extension's creator, Mikubill, the steps and UI may differ over time.
🐸This video covers the "Local Version AUTOMATIC1111," not the so-called "Colab Version," "One-Touch Version," or "NMKD Version."
🐸This video was created for technical research purposes. It is an example of how to use the software and is not intended as a recommendation for its use.
🟢What is Control-Net?
This technology originally referred to this:
https://github.com/lllyasviel/ControlNet
※Illyasviel's page on GitHub
(Updated April 2011)
Updated to version 1.1.
https://github.com/lllyasviel/Control...
🟢Generally, this refers to "sd-webui-controlnet," an extension implemented to run the above on AUTOMATIC1111.
https://github.com/Mikubill/sd-webui-...
※Mikubill's page on GitHub
🟢What can it do?
You can trace the line drawings of the original to create image differences, and easily create poses that were difficult to recreate using prompts alone.
So, what exactly is Stable Diffusion webUI AUTOMATIC1111? Click here if you're wondering how to install Stable Diffusion webUI AUTOMATIC1111.
🟠How to Install Stable Diffusion webUI AUTOMATIC1111
• 日本一わかりやすいStableDiffusion WebUI AUTOMATIC11...
↑ (As long as your computer meets the requirements) Follow the instructions in this video to install it.
🟠Author AUTOMATIC1111's Page
https://github.com/AUTOMATIC1111/stab...
🐸How to Install Control-Net (from 01:53)
[Note 1] If you have an older version of AUTOMATIC1111 (one that was installed in 2010 and is still being used), please update the version or perform a fresh install to get the latest version.
If the main program is outdated, installing ControlNet will not display any options. (My personal experience ('A`) Ugh!)
[Note 2] Before installing, please be sure to read the Github pages of the above authors (Illyasviel and Mikubiru).
*Added April 2011
The following steps have been updated with the v1.1 update.
1. Extensions tab → Install from URL tab → URL for extension's git Paste the following URL into the repository field and click the Install button.
https://github.com/Mikubill/sd-webui-...
② Obtaining the Estimation Model
https://huggingface.co/lllyasviel/Con...
Download the file with the .pth extension (units of gigabytes) from here (this is the estimation model).
Location for Estimation Model
Put it in the following location: installation_folder/extensions/sd-webui-controlnet/models/
(From v1.1, the .pth (estimation model) and .yaml (configuration file) are used as a set. Please make sure they have the same name.)
🐸About the Detect Map (06:03 - )
The extracted stick figure and line drawings are called detect maps.
In the default settings after installation, they are temporarily saved in a hidden system folder.
C:\Users\Username\AppData\Local\Temp (← for Windows 11)
You can open the folder and retrieve them each time, but honestly, it's a bit of a hassle, so I recommend setting it to auto-save.
Checking "Allow ~ Auto Saving" in the Settings tab will automatically save them to the location specified in the "Directory for detected maps auto saving" section above.
*By default, it's the "detected_maps" folder in the outputs folder.
↑This may vary depending on the installation date. Just check "Allow ~" and it should be somewhere in the installation folder, so look for it.
*Similar detect map images will accumulate each time you generate them, so if this bothers you, clean them up regularly.
🐸About the meaning of the checkbox settings
✅Enable
ControlNet on/off switch. Unlike scripts, it doesn't activate just by opening a tab, so be sure to check it when using it.
✅Invert Input Color
As written in English just above, check this box if the background color of the original image is white.
✅RGB to BGR: For normal maps (I honestly don't use this...)
✅Low VRAM: Check this box if your graphics card has low VRAM (for GTX16** users). This will slow down the generation speed.
✅Guess Mode: Guess mode. The author calls it a non-prompt mode.
This seems to be a test implementation, so it's for advanced users. Step count and other settings are specified.
https://github.com/Mikubill/sd-webui-...
🐸About the preprocessor and estimation model (05:11) ~)
Use the same preprocessor and estimation model.
Using different preprocessors will invalidate the settings.
If you want to recreate a pose from an image, use open pose for the preprocessor and open pose for the model.
If you want to recreate a pose from a stick figure image, since the stick figure has already been extracted, use preprocessor none (open pose will also work), and open pose for the model.
*Just because you're trying to recreate a pose doesn't mean y...