sd-webui-docker

Yet another Docker image for stable-diffusion-webui with Kubernetes support.

• Getting Started
• Usage
  • Run on GPU servers
  • CPU-only mode
  • Sync outputs to local
  • Mount your own models directory
  • Sync configs to local
• Kubernetes
  • Architecture
  • Deploy with Kustomize
  • Configure kustomization.yaml
    • Enable Ingress
    • Use Waifu Diffusion
• Acknowledgements
• Changelog

Getting Started

1. Prepare your work directory

# Go to your work dir first.
# E.g., `mkdir work && cd work`

# create dirs
mkdir -m 777 -p .cache embeddings extensions log models models/Stable-diffusion models/VAE-approx outputs

# create empty files
echo '{}' | tee config.json ui-config.json cache.json

2. Download models

# download Stable Diffusion v1.5 if not exists
DST=models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
if [ ! -f "$DST" ] ; then curl -L -o "$DST" https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors ; fi

# download VAE if not exists
DST=models/VAE-approx/model.pt
if [ ! -f "$DST" ] ; then curl -L -o "$DST" https://github.com/AUTOMATIC1111/stable-diffusion-webui/raw/master/models/VAE-approx/model.pt ; fi

3. Start the server

docker run --rm --gpus all -p 7860:7860 \
  -e TRANSFORMERS_CACHE=/work/stable-diffusion-webui/.cache \
  -v "$(pwd)"/.cache:/work/stable-diffusion-webui/.cache \
  -v "$(pwd)"/embeddings:/work/stable-diffusion-webui/embeddings \
  -v "$(pwd)"/extensions:/work/stable-diffusion-webui/extensions \
  -v "$(pwd)"/log:/work/stable-diffusion-webui/log \
  -v "$(pwd)"/models:/work/stable-diffusion-webui/models \
  -v "$(pwd)"/outputs:/work/stable-diffusion-webui/outputs \
  -v "$(pwd)"/config.json:/work/stable-diffusion-webui/config.json \
  -v "$(pwd)"/ui-config.json:/work/stable-diffusion-webui/ui-config.json \
  -v "$(pwd)"/cache.json:/work/stable-diffusion-webui/cache.json \
  ghcr.io/ebiiim/sd-webui --xformers --api

Then open http://localhost:7860 in your browser.

Usage

💡 Please read stable-diffusion-webui CLI docs before using this image.

Basically docker run --rm --gpus all -p 7860:7860 <IMAGE> [ARGS].... The entrypoint is launch.py --listen so please pass additional arguments.

Run on GPU servers

Optionally pass --xformers --api.

docker run --rm --gpus all -p 7860:7860 ghcr.io/ebiiim/sd-webui --xformers --api

CPU-only mode

Pass --skip-torch-cuda-test --no-half --use-cpu all, and optionally pass --api.

docker run --rm -p 7860:7860 ghcr.io/ebiiim/sd-webui --skip-torch-cuda-test --no-half --use-cpu all --api

Sync outputs to local

Use Docker -v flag.

mkdir -p "$(pwd)"/outputs
docker run --rm --gpus all -p 7860:7860 \
  -v "$(pwd)"/outputs:/work/stable-diffusion-webui/outputs \
  ghcr.io/ebiiim/sd-webui \
  --xformers --api

Mount your own models directory

Use Docker -v flag.

MODELS_DIR=/path/to/models
docker run --rm --gpus all -p 7860:7860 \
  -v "${MODELS_DIR}":/work/stable-diffusion-webui/models \
  ghcr.io/ebiiim/sd-webui \
  --xformers --api

Sync configs to local

Use Docker -v flag.

echo '{}' | tee config.json ui-config.json cache.json
docker run --rm --gpus all -p 7860:7860 \
  -v "$(pwd)"/config.json:/work/stable-diffusion-webui/config.json \
  -v "$(pwd)"/ui-config.json:/work/stable-diffusion-webui/ui-config.json \
  -v "$(pwd)"/cache.json:/work/stable-diffusion-webui/cache.json \
  ghcr.io/ebiiim/sd-webui \
  --xformers --api

Kubernetes

Architecture

⚠️ This is a bit old and now we have more PVCs to sync other resources.

Deploy with Kustomize

💡 Please make sure GPUs are enabled on your cluster.

Create the namespace,

kubectl create ns stable-diffusion-webui

apply manifests,

kubectl apply -k https://github.com/ebiiim/sd-webui-docker/k8s

forward the port to access,

kubectl port-forward -n stable-diffusion-webui svc/stable-diffusion-webui 7860:7860

and open http://localhost:7860 in your browser.

Configure kustomization.yaml

Download kustomization.yaml and edit it.

curl -Lo https://raw.githubusercontent.com/ebiiim/sd-webui-docker/main/k8s/kustomization.yaml

Enable Ingress

Uncomment the resource,

 resources:
   # Ingress (optional)
-  # - https://raw.githubusercontent.com/ebiiim/sd-webui-docker/v1.2.0/k8s/bases/ing.yaml
+  - https://raw.githubusercontent.com/ebiiim/sd-webui-docker/v1.2.0/k8s/bases/ing.yaml

and set your domain name.

 patches:
   # [Ingress] To set ingressClassName and host name, uncomment the patch.
   - target:
       group: networking.k8s.io
       version: v1
       kind: Ingress
       name: stable-diffusion-webui
     patch: |-
       # - op: add
       #   path: /spec/ingressClassName
       #   value: nginx
-      # - op: replace
-      #   path: /spec/rules/0/host
-      #   value: stable-diffusion.localdomain
+      - op: replace
+        path: /spec/rules/0/host
+        value: [YOUR_DOMAIN_HERE]

Use Waifu Diffusion

Uncomment the resource.

  resources:
    # [MODELS] Please install 1 or more models.
    # Stable Diffusion 1.5 (License: CreativeML Open RAIL-M)
    - https://raw.githubusercontent.com/ebiiim/sd-webui-docker/v1.2.0/k8s/models/install-sd15.yaml
    # Waifu Diffusion 1.5 beta2 (License: Fair AI Public License 1.0-SD)
-   # - https://raw.githubusercontent.com/ebiiim/sd-webui-docker/v1.2.0/k8s/models/install-wd15b2.yaml
+   - https://raw.githubusercontent.com/ebiiim/sd-webui-docker/v1.2.0/k8s/models/install-wd15b2.yaml

Acknowledgements

This work is based on projects whose licenses are listed below.

• AUTOMATIC1111/stable-diffusion-webui
  • https://github.com/AUTOMATIC1111/stable-diffusion-webui#credits
  • https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/LICENSE.txt
• NVIDIA Deep Learning Container
  • https://hub.docker.com/r/nvidia/cuda
  • https://developer.nvidia.com/ngc/nvidia-deep-learning-container-license

Changelog

1.2.0 - 2023-03-25

• update docs
• K8s: add PVCs to sync resources

1.1.0 - 2023-03-22

• Kubernetes support

1.0.0 - 2023-03-18

• initial release
• stable-diffusion-webui: a9fed7c364061ae6efb37f797b6b522cb3cf7aa2 2023-03-14