Windows+AMD support has not officially been made for webui,
but you can install lshqqytiger's fork of webui that uses Direct-ml.

-Training currently doesn't work, yet a variety of features/extensions do, such as LoRAs and controlnet. Report issues at https://github.com/lshqqytiger/stable-diffusion-webui-directml/issues

1. Install Python 3.10.6 (ticking Add to PATH), and git
2. paste this line in cmd/terminal: git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
   ^{(you can move the program folder somewhere else.)}
3. Double-click webui-user.bat
4. If it looks like it is stuck when installing or running, press enter in the terminal and it should continue.

(The rest below are installation guides for linux with rocm.)

^{(As of 1/15/23 you can just run webui.sh and pytorch+rocm should be automatically installed for you.)}

1. Enter these commands, which will install webui to your current directory:

sudo apt install git python3.10-venv -y
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui && cd stable-diffusion-webui
python3.10 -m venv venv

2. Install and run with:

   ./webui.sh {your_arguments*}

*For many AMD GPUs, you must add --precision full --no-half or --upcast-sampling arguments to avoid NaN errors or crashing. If --upcast-sampling works as a fix with your card, you should have 2x speed (fp16) compared to running in full precision.

Execute the following:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui
python -m venv venv
source venv/bin/activate
python -m pip install --upgrade pip wheel

# It's possible that you don't need "--precision full", dropping "--no-half" however crashes my drivers
TORCH_COMMAND='pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/rocm5.1.1' python launch.py --precision full --no-half

In following runs you will only need to execute:

cd stable-diffusion-webui
# Optional: "git pull" to update the repository
source venv/bin/activate

# It's possible that you don't need "--precision full", dropping "--no-half" however crashes my drivers
TORCH_COMMAND='pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/rocm5.1.1' python launch.py --precision full --no-half

The first generation after starting the WebUI might take very long, and you might see a message similar to this:

MIOpen(HIP): Warning [SQLiteBase] Missing system database file: gfx1030_40.kdb Performance may degrade. Please follow instructions to install: https://github.com/ROCmSoftwarePlatform/MIOpen#installing-miopen-kernels-package

The next generations should work with regular performance. You can follow the link in the message, and if you happen to use the same operating system, follow the steps there to fix this issue. If there is no clear way to compile or install the MIOpen kernels for your operating system, consider following the "Running inside Docker"-guide below.

Pull the latest rocm/pytorch Docker image, start the image and attach to the container (taken from the rocm/pytorch documentation): docker run -it --network=host --device=/dev/kfd --device=/dev/dri --group-add=video --ipc=host --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v $HOME/dockerx:/dockerx rocm/pytorch

Execute the following inside the container:

cd /dockerx
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui
python -m pip install --upgrade pip wheel

# It's possible that you don't need "--precision full", dropping "--no-half" however crashes my drivers
REQS_FILE='requirements.txt' python launch.py --precision full --no-half

Following runs will only require you to restart the container, attach to it again and execute the following inside the container: Find the container name from this listing: docker container ls --all, select the one matching the rocm/pytorch image, restart it: docker container restart <container-id> then attach to it: docker exec -it <container-id> bash.

cd /dockerx/stable-diffusion-webui
# Optional: "git pull" to update the repository

# It's possible that you don't need "--precision full", dropping "--no-half" however crashes my drivers
REQS_FILE='requirements.txt' python launch.py --precision full --no-half

The /dockerx folder inside the container should be accessible in your home directory under the same name.

If the web UI becomes incompatible with the pre-installed Python 3.7 version inside the Docker image, here are instructions on how to update it (assuming you have successfully followed "Running inside Docker"):

Execute the following inside the container:

apt install python3.9-full # Confirm every prompt
update-alternatives --install /usr/local/bin/python python /usr/bin/python3.9 1
echo 'PATH=/usr/local/bin:$PATH' >> ~/.bashrc

Run source ~/.bashrc and proceed by running the same commands as you would with an existing container.

It's possible that you don't need "--precision full", dropping "--no-half" however it may not work for everyone. Certain cards like the Radeon RX 6000 Series and the RX 500 Series will function normally without the option --precision full --no-half, saving plenty of VRAM. (noted here.)

Install webui on Arch Linux with Arch-specific packages
and possibly other Arch-based Linux distributions (tested Feb 22 2023)

1. Start with required dependencies and install pip

sudo pacman -S python-pip

2. Install pytorch with ROCm backend

Arch [Community] repository offers two pytorch packages, python-pytorch-rocm and python-pytorch-opt-rocm. For CPUs with AVX2 instruction set support, that is, CPU microarchitectures beyond Haswell (Intel, 2013) or Excavator (AMD, 2015), install python-pytorch-opt-rocm to benefit from performance optimizations. Otherwise install python-pytorch-rocm:

# Install either one:
sudo pacman -S python-pytorch-rocm
sudo pacman -S python-pytorch-opt-rocm   # AVX2 CPUs only

3. Install torchvision with ROCm backend

python-torchvision-rocm package is located in AUR. Clone the git repository and compile the package on your machine

git clone https://aur.archlinux.org/python-torchvision-rocm.git
cd python-torchvision-rocm
makepkg -si

Confirm all steps until Pacman finishes installing python-torchvision-rocm.

Alternatively, install the python-torchvision-rocm package with a AUR helper.

1. Manually create a venv environment with system site-packages (this will allows access to system pytorch and torchvision). Install the remaining Python dependencies

python -m venv venv --system-site-packages
source venv/bin/activate
pip install -r requirements.txt

Run the following inside the project root to start webui:

source venv/bin/activate
./webui.sh

Depending on the GPU model, you may need to add certain Command Line Arguments and Optimizations to webui-user.sh in order for webui to run properly. Refer to the Automatic Installation section.

• GPU model has to be supported by Arch dependencies

See if your GPU is listed as a build architecture in PYTORCH_ROCM_ARCH variable for Tourchvision and PyTorch. References for architectures can be found here. If not, consider building both packages locally or use another installation method.

• Arch dependencies (pytorch, torchvision) are kept up-to-date by full system updates (pacman -Syu) and compiling, which may not be desirable when dependency combinations with fixed versions are wished

This guide has been tested on AMD Radeon RX6800 with Python 3.10.9, ROCm 5.4.3, PyTorch 1.13.1, Torchvision 0.14.1