How to Deploy a CPU‑Based Stable Diffusion Service on Alibaba Cloud ACK

Prerequisites

A pre‑created Alibaba Cloud Container Service (ACK) Kubernetes managed cluster. CPU deployment requires a node with at least 8 CPU × 16 GB RAM; GPU deployment requires a GPU node with >40 GB free disk.

kubectl configured to access the cluster.

Stable Diffusion Docker image hosted at

zibai-registry.cn-hangzhou.cr.aliyuncs.com/gpt/stable-diffusion

(CPU tag v1.cpu, GPU tag v1.gpu).

CPU version deployment via console

Log in to the Container Service Management Console and select the target cluster.

Navigate to Workloads → Stateless and click Create from Image .

Enter basic application information (name, namespace, etc.) and confirm.

After the workload is created, go to Network → Services and create a Service of type LoadBalancer .

Wait ~1 minute for an external IP to be allocated, then refresh the service page to view the External IP column.

Access the UI at http://<ExternalIP>:7860.

CPU version deployment via kubectl

apiVersion: apps/v1
kind: Deployment
metadata:
  name: stable-diffusion
  namespace: default
  labels:
    app: stable-diffusion
spec:
  replicas: 1
  selector:
    matchLabels:
      app: stable-diffusion
  template:
    metadata:
      labels:
        app: stable-diffusion
    spec:
      containers:
      - name: stable-diffusion
        image: zibai-registry.cn-hangzhou.cr.aliyuncs.com/gpt/stable-diffusion:v1.cpu
        imagePullPolicy: IfNotPresent
        command: ["python3", "launch.py"]
        args: ["--listen", "--skip-torch-cuda-test", "--no-half"]
        resources:
          requests:
            cpu: "2"
            memory: 2Gi
---
apiVersion: v1
kind: Service
metadata:
  name: stable-diffusion
  namespace: default
  annotations:
    service.beta.kubernetes.io/alibaba-cloud-loadbalancer-address-type: internet
    service.beta.kubernetes.io/alibaba-cloud-loadbalancer-instance-charge-type: PayByCLCU
spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
  ports:
  - port: 7860
    protocol: TCP
    targetPort: 7860
  selector:
    app: stable-diffusion

kubectl apply -f stable-diffusion.yaml

Monitor pod status with kubectl get po | grep stable-diffusion and retrieve the load balancer IP with kubectl get svc stable-diffusion. The CPU image is ~12.7 GB and typically downloads in ~10 minutes.

GPU version deployment

Create a GPU‑enabled ACK cluster (node must have a GPU and >40 GB free disk).

apiVersion: apps/v1
kind: Deployment
metadata:
  name: stable-diffusion
  namespace: default
  labels:
    app: stable-diffusion
spec:
  replicas: 1
  selector:
    matchLabels:
      app: stable-diffusion
  template:
    metadata:
      labels:
        app: stable-diffusion
    spec:
      containers:
      - name: stable-diffusion
        image: zibai-registry.cn-hangzhou.cr.aliyuncs.com/gpt/stable-diffusion:v1.gpu
        imagePullPolicy: IfNotPresent
        command: ["python3", "launch.py"]
        args: ["--listen", "--skip-torch-cuda-test", "--no-half"]
        resources:
          requests:
            cpu: "2"
            memory: 2Gi
          limits:
            nvidia.com/gpu: 1
---
apiVersion: v1
kind: Service
metadata:
  name: stable-diffusion
  namespace: default
  annotations:
    service.beta.kubernetes.io/alibaba-cloud-loadbalancer-address-type: internet
    service.beta.kubernetes.io/alibaba-cloud-loadbalancer-instance-charge-type: PayByCLCU
spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
  ports:
  - port: 7860
    protocol: TCP
    targetPort: 7860
  selector:
    app: stable-diffusion

kubectl apply -f stable-diffusion.yaml

After the pod becomes ready, the GPU image (~15.1 GB) typically downloads in ~15 minutes. Access the UI at the same http://<ExternalIP>:7860; image generation is noticeably faster than the CPU version.

Notes

Image pull times are measured up to 2023‑05‑17.

Image repository:

zibai-registry.cn-hangzhou.cr.aliyuncs.com/gpt/stable-diffusion

.

Reference implementation: https://github.com/AUTOMATIC1111/stable-diffusion-webui