部署Stable Diffusion-Elastic Compute Service(ECS)-阿里云帮助中心

Background

Stable Diffusion is a text-to-image latent diffusion model that generates images from text prompts. It works by progressively denoising a random Gaussian noise input to create a target image. Compared to traditional generative models, Stable Diffusion can produce high-fidelity, detailed images without requiring complex training processes or massive datasets. It is now widely used in various fields, including computer vision, digital art, and video games.

Create an ECS instance

Go to the instance creation page.
On the instance creation page, configure the parameters to create an ECS instance.

Configure the following key parameters. For information about other parameters, see Create an instance by using the wizard.
- Instance: Stable Diffusion model inference is memory-intensive. To ensure stable operation, select at least the ecs.g8a.16xlarge instance type (256 GiB of memory).
- Images: Alibaba Cloud Linux 3.2104 LTS 64-bit.
- Public IP Address: Select Assign Public IPv4 Address. Set the bandwidth billing method to Pay-by-traffic and the peak bandwidth to 100 Mbps. This helps accelerate model downloads.
- System Disk: Running Stable Diffusion requires significant storage space for model files. To ensure smooth operation, set the system disk size to at least 100 GiB.
Add security group rules.

Add an inbound rule to the security group of the ECS instance to allow traffic on ports 22, 443, and 7860 (for WebUI access). For more information, see Add a security group rule.
After the instance is created, obtain its public IP address from the ECS instance page.

Note
Use the public IP address to access the WebUI service and generate images.

Create the Docker runtime environment

Remotely connect to the ECS instance.

For more information, see Connect to a Linux instance by using Workbench.
Install Docker.

For more information, see Install Docker on an Alibaba Cloud Linux 3 instance.
Create and run the PyTorch AI container.

The Anolis community provides a rich selection of container images based on Anolis OS, including a PyTorch image optimized for AMD. You can use this image to create a PyTorch runtime environment.

The following commands pull the container image, create a container named pytorch-amd that runs in detached mode, and map your home directory to the container to preserve your development files.
```
sudo docker pull registry.openanolis.cn/openanolis/pytorch-amd:1.13.1-23-zendnn4.1
sudo docker run -d -it --name pytorch-amd --net host -v $HOME:/root registry.openanolis.cn/openanolis/pytorch-amd:1.13.1-23-zendnn4.1
```

Deploy Stable Diffusion

Manual

Step 1: Install required software

Enter the container environment.
```
sudo docker exec -it -w /root pytorch-amd /bin/bash
```
Important
Subsequent commands must be run in the container environment. If you exit unexpectedly, run the preceding command to re-enter. To check if you are in a container, run cat /proc/1/cgroup | grep docker. If the command returns output, you are in the container.

Install the software required for deploying Stable Diffusion.

yum install -y git git-lfs wget tmux mesa-libGL gperftools-libs

Enable Git LFS.

Git LFS is required to download the pre-trained models.
```
git lfs install
```

Step 2: Download source code and models

Create a tmux session.
```
tmux
```
Note
Downloading the pre-trained models can be time-consuming and its success depends on network conditions. Run the download in a tmux session to prevent interruptions from a lost connection.

Download the Stable Diffusion WebUI project source code.

git clone -b v1.5.2 https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

Create a repositories directory and download the project's dependencies.

mkdir stable-diffusion-webui/repositories && cd $_
git clone https://github.com/Stability-AI/stablediffusion.git stable-diffusion-stability-ai
git clone https://github.com/Stability-AI/generative-models.git generative-models
git clone https://github.com/crowsonkb/k-diffusion.git k-diffusion
git clone https://github.com/sczhou/CodeFormer.git CodeFormer
git clone https://github.com/salesforce/BLIP.git BLIP

Check out the specific commits for the dependency repositories to ensure stable generation results.

git -C stable-diffusion-stability-ai checkout cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf
git -C generative-models checkout 5c10deee76adad0032b412294130090932317a87
git -C k-diffusion checkout c9fe758757e022f05ca5a53fa8fac28889e4f1cf
git -C CodeFormer checkout c5b4593074ba6214284d6acd5f1719b6c5d739af
git -C BLIP checkout 48211a1594f1321b00f14c9f7a5b4813144b2fb9

Download the Stable Diffusion pre-trained model.

cd ~ && mkdir -p stable-diffusion-webui/models/Stable-diffusion
wget "https://www.modelscope.cn/models/AI-ModelScope/stable-diffusion-v1-5/resolve/master/v1-5-pruned-emaonly.safetensors" -O stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
mkdir -p ~/stable-diffusion-webui/models/clip
git clone --depth=1 https://gitee.com/modelee/clip-vit-large-patch14.git ~/stable-diffusion-webui/models/clip/clip-vit-large-patch14

Modify the script so that Stable Diffusion uses the local ViT multimodal model at runtime.

sed -i "s?openai/clip-vit-large-patch14?${HOME}/stable-diffusion-webui/models/clip/clip-vit-large-patch14?g" ~/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/modules/encoders/modules.py

Step 3: Deploy runtime environment

Change the pip download source.

Before you install dependencies, change the pip download source to accelerate the installation.
1. Create the pip directory.
```
mkdir -p ~/.config/pip
```
2. Configure the pip installation mirror.
```
cat > ~/.config/pip/pip.conf <<EOF
[global]
index-url=http://mirrors.cloud.aliyuncs.com/pypi/simple/
[install]
trusted-host=mirrors.cloud.aliyuncs.com
EOF
```

Install Python runtime dependencies.

pip install cython gfpgan open-clip-torch==2.8.0 httpx==0.24.1
pip install git+https://github.com/openai/CLIP.git@d50d76daa670286dd6cacf3bcd80b5e4823fc8e1

Set the OMP_NUM_THREADS and GOMP_CPU_AFFINITY environment variables.

Set the OMP_NUM_THREADS and GOMP_CPU_AFFINITY environment variables, which are required by the ZenDNN runtime library for your hardware platform.
```
cat > /etc/profile.d/env.sh <<EOF
export OMP_NUM_THREADS=\$(nproc --all)
export GOMP_CPU_AFFINITY=0-\$(( \$(nproc --all) - 1 ))
EOF
source /etc/profile
```

Run the script to automatically deploy the Stable Diffusion runtime environment.

cd ~/stable-diffusion-webui
venv_dir="-" ./webui.sh -f --skip-torch-cuda-test --exit

Step 4: Generate images

Run the following command to start the WebUI service.

export LD_PRELOAD=/usr/lib64/libtcmalloc.so.4
export venv_dir="-" 
python3 launch.py -f --skip-torch-cuda-test --skip-version-check --no-half --precision full --use-cpu all --listen

The WebUI service has started successfully when the following output appears:

Python 3.10.12 (main, Jun  7 2023, 00:00:00) [GCC 12.2.1 20221121 (Anolis OS 12.2.1-2)]
Version: v1.5.2
Commit hash: c9c8485bc1e8720aba70f029d25cba1c4abf2b5c
Launching Web UI with arguments: -f --skip-torch-cuda-test --skip-version-check --no-half --precision full --use-cpu all --listen
no module 'xformers'. Processing without...
No SDP backend available, likely because you are running in pytorch versions < 2.0. In fact, you are in PyTorch 1.13.1. You might want to consider upgrading.
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled
Calculating sha256 for /root/stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors: Running on local URL:  http://0.0.0.0:7860
To create a public link, set `share=True` in `launch()`.
Startup time: 4.1s (launcher: 0.2s, import torch: 1.7s, import gradio: 0.6s, setup paths: 0.4s, other imports: 0.5s, load scripts: 0.3s, create ui: 0.3s, gradio launch: 0.1s).
6ce0161689b3853acaa03779ec93eafe75a02f4ced659bee03f50797806fa2fa
Loading weights [6ce0161689] from /root/stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
Creating model from config: /root/stable-diffusion-webui/configs/v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Applying attention optimization: InvokeAI... done.
Model loaded in 3.8s (calculate hash: 2.8s, load config: 0.1s, create model: 0.3s, apply weights to model: 0.3s, calculate empty prompt: 0.1s).

In your browser's address bar, enter http://<ECS_public_IP_address>:7860 to access the web interface.
In the prompt text box, enter a prompt (English only), for example, Urban portrait of a skateboarder in mid-jump, graffiti walls background, high shutter speed. Click Generate to create the image.

Note
You can also try other prompts to explore more capabilities of Stable Diffusion.

Automated

Alibaba Cloud provides an automated deployment script for deploying and running the Stable Diffusion model with a single command.

Download the automated deployment script.

wget https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/en-US/20231213/bild/deploy_stable-diffusion_amd-docker.sh

Enter the container environment.
```
sudo docker exec -it -w /root pytorch-amd /bin/bash
```
Important
Subsequent commands must be run in the container environment. If you exit unexpectedly, run the preceding command to re-enter. To check if you are in a container, run cat /proc/1/cgroup | grep docker. If the command returns output, you are in the container.
Install tmux and create a tmux session.
```
yum install -y tmux
tmux
```
Note
This script downloads model files and can take a long time to run. Run it in a tmux session to prevent interruptions from a lost connection.
Make the automated deployment script executable.
```
chmod +x deploy_stable-diffusion_amd-docker.sh
```

Run the automated deployment script.

./deploy_stable-diffusion_amd-docker.sh

The automated deployment is complete when the following output appears:

##############################################################
#                                                            #
#        Deploy success! Service is running now.             #
#                                                            #
##############################################################
[    @iZuf6fwy177fk    xxx    ~]# nohup: ignoring input and redirecting stderr to stdout
Python 3.10.12 (main, Jun  7 2023, 00:00:00) [GCC 12.2.1 20221121 (Anolis OS 12.2.1-2)]
Version: v1.5.2
Commit hash: c9c8485bc1e8720aba70f029d25cba1c4abf2b5c
Launching Web UI with arguments: -f --skip-torch-cuda-test --skip-version-check --no-half --precision full --use-cpu all --listen
no module 'xformers'. Processing without...
No SDP backend available, likely because you are running in pytorch versions < 2.0. In fact, you are using PyTorch 1.13.1. You might want to consider upgrading.
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled
Calculating sha256 for /root/stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors: Running on local URL:  http://0.0.0.0:7860
To create a public link, set `share=True` in `launch()`.
Startup time: 4.7s (launcher: 0.2s, import torch: 2.2s, import gradio: 0.6s, setup paths: 0.4s, other imports: 0.4s, setup codeformer: 0.2s, load scripts: 0.3s, create ui: 0.3s, gradio launch: 0.1s).
6ce0161689b3853acaa03779ec93eafe75a02f4ced659bee03f50797806fa2fa
Loading weights [6ce0161689] from /root/stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
Creating model from config: /root/stable-diffusion-webui/configs/v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Applying attention optimization: InvokeAI... done.
Model loaded in 3.8s (calculate hash: 2.8s, create model: 0.4s, apply weights to model: 0.3s, calculate empty prompt: 0.1s).

Note

The script takes a long time to run. If you lose your connection to the ECS instance during deployment in a tmux session, you can reconnect to the instance and run tmux attach to resume the session and check the script's progress.

In your browser's address bar, enter http://<ECS_public_IP_address>:7860 to access the web interface.
In the prompt text box, enter a prompt (English only), for example, Urban portrait of a skateboarder in mid-jump, graffiti walls background, high shutter speed. Click Generate to create the image.

Note
You can also try other prompts to explore more capabilities of Stable Diffusion.