Container Compute Service:Accelerate Wan2.1 video generation with DeepGPU

Background

Prerequisites

Before you begin, ensure that you have:

• ACS account authorization. If this is your first time using ACS, assign the default role so ACS can access Elastic Compute Service (ECS), Object Storage Service (OSS), Apsara File Storage NAS, Cloud Parallel File Storage (CPFS), and Server Load Balancer (SLB). For details, see Quick start for first-time ACS users

• An ACS GPU cluster with L20 (GN8IS) or G49E GPU cards

• A NAS or OSS persistent volume to store model files. This guide uses a NAS volume. For setup instructions, see Create a NAS file system as a volume or Use a statically provisioned OSS volume

• Git installed in your local environment. See Git downloads

Step 1: Prepare the model data

Run the following commands in the directory where the NAS volume is mounted.

1. Clone the ComfyUI repository.

   git clone https://github.com/comfyanonymous/ComfyUI.git

2. Download the three Wan2.1 model files to their corresponding ComfyUI directories. The files are hosted on the Wan_2.1_ComfyUI_repackaged project on ModelScope.

   • The diffusion model (wan2.1_t2v_14B_fp16.safetensors):

     cd ComfyUI/models/diffusion_models
     wget https://modelscope.cn/models/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/master/split_files/diffusion_models/wan2.1_t2v_14B_fp16.safetensors

   • The VAE (wan_2.1_vae.safetensors):

     cd ComfyUI/models/vae
     wget https://modelscope.cn/models/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/master/split_files/vae/wan_2.1_vae.safetensors

   • The text encoder (umt5_xxl_fp8_e4m3fn_scaled.safetensors):

     cd ComfyUI/models/text_encoders
     wget https://modelscope.cn/models/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/master/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors

   The download takes around 30 minutes. If your connection is slow, increase the peak public bandwidth before starting.

3. Download and extract the ComfyUI-deepgpu plugin.

   cd ComfyUI/custom_nodes
   wget https://aiacc-inference-public-v2.oss-cn-hangzhou.aliyuncs.com/deepgpu/comfyui/nodes/20250513/ComfyUI-deepgpu.tar.gz
   tar zxf ComfyUI-deepgpu.tar.gz

Step 2: Deploy the ComfyUI service

1. Log in to the ACS console. In the left navigation pane, choose Clusters. Click the target cluster name. Then choose Workloads > Deployments and click Create from YAML.

2. Paste the following YAML manifest and click Create.

   Replace persistentVolumeClaim.claimName with the name of your persistent volume claim (PVC). This example uses the inference-nv-pytorch 25.07 image from the cn-beijing region to minimize image pull times. To use this image from other regions, see Usage method and update the image path in the manifest. The container image in this example has the deepgpu-torch and deepgpu-comfyui plugins pre-installed. To use these plugins in a different container environment, contact a solution architect (SA) to get the installation packages.

   apiVersion: apps/v1
   kind: Deployment
   metadata:
     labels:
       app: wanx-deployment
     name: wanx-deployment-test
     namespace: default
   spec:
     replicas: 1
     selector:
       matchLabels:
         app: wanx-deployment
     template:
       metadata:
         labels:
           alibabacloud.com/compute-class: gpu
           alibabacloud.com/compute-qos: default
           alibabacloud.com/gpu-model-series: L20 #Supported GPU card types: L20 (GN8IS), G49E
           app: wanx-deployment
       spec:
         containers:
         - command:
           - sh
           - -c
           - DEEPGPU_PUB_LS=true python3 /mnt/ComfyUI/main.py --listen 0.0.0.0 --port 7860
           image: acs-registry-vpc.cn-beijing.cr.aliyuncs.com/egslingjun/inference-nv-pytorch:25.07-vllm0.9.2-pytorch2.7-cu128-20250714-serverless
           imagePullPolicy: Always
           name: main
           resources:
             limits:
               nvidia.com/gpu: "1"
               cpu: "16"
               memory: 64Gi
             requests:
               nvidia.com/gpu: "1"
               cpu: "16"
               memory: 64Gi
           terminationMessagePath: /dev/termination-log
           terminationMessagePolicy: File
           volumeMounts:
           - mountPath: /dev/shm
             name: cache-volume
           - mountPath: /mnt #/mnt is the path in the pod where the NAS volume claim is mapped
             name: data
         dnsPolicy: ClusterFirst
         restartPolicy: Always
         schedulerName: default-scheduler
         securityContext: {}
         terminationGracePeriodSeconds: 30
         volumes:
         - emptyDir:
             medium: Memory
             sizeLimit: 500G
           name: cache-volume
         - name: data
           persistentVolumeClaim:
             claimName: wanx-nas #wanx-nas is the volume claim created from the NAS volume
   ---
   apiVersion: v1
   kind: Service
   metadata:
     name: wanx-test
   spec:
     type: LoadBalancer
     ports:
       - port: 7860
         protocol: TCP
         targetPort: 7860
     selector:
       app: wanx-deployment

   Key parameters in this manifest:

   Parameter	Description
   alibabacloud.com/gpu-model-series	GPU card type. Supported values: L20 (GN8IS instance) and G49E.
   nvidia.com/gpu: "1"	Requests one GPU for the container.
   resources.limits/requests	Sets CPU to 16 cores and memory to 64 GiB.
   /dev/shm emptyDir (sizeLimit: 500G)	Shared memory volume mounted at /dev/shm. Required for large model inference.
   mountPath: /mnt	The path inside the pod where the NAS volume is mounted. ComfyUI and model files are accessed from this path.
   persistentVolumeClaim.claimName	Name of your PVC. Replace wanx-nas with your actual PVC name.

3. In the dialog that appears, click View to open the workload details page. Click the Logs tab. When the service starts successfully, the log output looks like this:

   

Step 3: Access the ComfyUI interface

1. On the workload details page, click the Access Method tab to get the external endpoint of the service, such as 8.xxx.xxx.114:7860.

   

2. Open http://8.xxx.xxx.114:7860/ in a browser.

   The first time you access the URL, it may take about 5 minutes to load.

3. In the ComfyUI interface, right-click anywhere and click Add Node to browse the available DeepGPU nodes from the plugin. The ApplyDeepyTorch node optimizes diffusion model inference by applying GPU-level acceleration. Insert it after the last model-loading node in your workflow. The node looks like this:

   

   

Step 4: Run the accelerated workflow

Download one or both of the following pre-built Wan2.1 workflows to your local machine:

• Image-to-video workflow: https://aiacc-inference-public-v2.oss-cn-hangzhou.aliyuncs.com/deepgpu/comfyui/wan/workflows/workflow_image_to_video_wan_1.3b_deepytorch.json

• Text-to-video workflow: https://aiacc-inference-public-v2.oss-cn-hangzhou.aliyuncs.com/deepgpu/comfyui/wan/workflows/workflow_text_to_video_wan_deepytorch.json

The following steps use the accelerated text-to-video workflow as an example.

1. In ComfyUI, choose Workflow > Open and select the downloaded workflow_text_to_video_wan_deepytorch.json file.

2. Find the Apply DeepyTorch to diffusion model node. Set its enable parameter to true.

   The DeepyTorch-accelerated workflow inserts an ApplyDeepyTorch node after the Load Diffusion Model node.

   

3. Click Run and wait for the video to generate.

4. Click the Queue button on the left to view the generation time and preview the output.

   The first run takes longer than subsequent runs as the model warms up. Run the workflow two or three more times to see stable performance.

   

5. (Optional) To compare generation time without acceleration, restart the ComfyUI service and run the non-accelerated workflow: https://aiacc-inference-public-v2.oss-cn-hangzhou.aliyuncs.com/deepgpu/comfyui/wan/workflows/workflow_text_to_video_wan.json