1. Configure the Python environment
1. Install conda
Script acquisition link: https://repo.anaconda.com/archive/
Press Enter all the way, the input yes input yes
wget -c https://repo.anaconda.com/archive/Anaconda3-2023.03-1-Linux-x86_64.sh
2. Install python3.10.9 environment
Use conda to install python3.10.9 environment
- If you have not installed conda, install conda first, or search for Linux to install python3.10.9 for installation
conda create -n sd python=3.10.9 -c conda-forge -y
After the creation is complete, use the environment
conda activate sd
2. Install PyTorch
1. CPU version
Official website: https://pytorch.org/get-started/locally/#mac-prerequisites
Both Linux and Mac can use the following download method:
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu
2. If there is a GPU, check the cuda version first
Execute the command nvidia-smi to check whether the cuda version in the upper right corner is greater than or equal to 11.8. If it is greater than or equal to 11.8, execute the following command to install PyTorch
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
2.1 Install NVIDIA driver
- Reinstall graphics driver
- Confirm the model of the graphics card and execute the command: lspci
- Enter the nvidia official website to download the corresponding driver: https://www.nvidia.cn/geforce/drivers/
Uninstall the original driver
sudo apt remove --purge "nvidia-*" -y
install driver
sudo ./NVIDIA-Linux-x86_64-530.41.03.run -no-x-check -no-nouveau-check -no-opengl-files //安装
2.3 Install CUDA
CUDA official website download address: https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_network
apt-get install libnvidia-decode-530
apt-get install libnvidia-encode-530
apt-get install nvidia-driver-530
apt install nvidia-settings nvidia-prime libnvidia-compute-530:i386 libnvidia-decode-530:i386 libnvidia-encode-530:i386 libnvidia-fbc1-530:i386 libnvidia-gl-530:i386
Modify environment variables
echo 'export PATH="/usr/local/cuda-12/bin:$PATH"' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH="/usr/local/cuda-12/LIB64:$LD_LIBRARY_PATH"' >> ~/.bashrc
source ~/.bashrc
nvcc -V
2.4 Error reporting when installing NVIDIA or CUDA
一、NVIDIA driver install - Error: Unable to find the kernel source tree
Solution
sudo apt-get install linux-headers-`uname -r`
2. Error Cannot locate TCMalloc
Solution
sudo apt-get install libgoogle-perftools4 libtcmalloc-minimal4 -y
3. Install the required components in advance
Replace the Tsinghua source first, and the download will be faster (it doesn’t matter if you hang an agent)
pip3 config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
Download components
pip3 install gfpgan ftfy regex tqdm
pip install git+https://github.com/openai/CLIP.git
Three, stable diffusion start
3.1 Installation
From the clone source code on GitHub, it is recommended to hang the proxy
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
Enter the directory, start the installation and start
cd stable-diffusion-webui
./webui.sh
3.2 Start mode
MAC needs to execute the following startup commands
./webui.sh --no-half --enable-insecure-extension-access
modify port
- Method 1: Edit webui-user.sh, the content is as follows, and then execute ./wehui.sh to start
# 绑定 0.0.0.0 端口，同时修改端口
export COMMANDLINE_ARGS="--listen --port 23105"
- Method 2: Add parameters at startup
./webui.sh --enable-insecure-extension-access --disable-safe-npickle --listen --port 20022
4. Replace the extension URL
https://gitee.com/akegarasu/sd-webui-extensions/raw/master/index.md
5. Sinicization course
1. Installation
Extensions–>install from URL, enter the Chinese plug-in address, click install
https://github.com/dtlnor/stable-diffusion-webui-localization-zh_CN
2. Select language pack
settings–>User interface, click the refresh button, select zh_CN language pack
3. Submit settings
First click Apply settings to submit the settings, then click Reload UI to restart the interface.
6. Plug-ins
1. Multi Diffusion + Tiled VAE
A powerful plug-in that allows players with 4G and 6G low video memory to generate 2K, 4K or even 8K pictures
The principle is to split the original image into small pieces, and finally aggregate it into a large image. The larger the block, the slower it will be, but it will be more natural
https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
2. inpaint cutout
3. TILE model
7. Prompt words
1. Improve the success rate of hands
hand of guido daniele
8. Frequently asked questions
1. The problem of generating image garbled characters
Sampled using: DDIM, PLMS, UniPC