Starting a Pod

Use this guide to choose a ComfyUI template, deploy it on RunPod and confirm that the pod starts correctly. Deployment usually takes several minutes because RunPod must download the container and the startup script may also download models from Docker Hub and the Hugging Face Hub.

If RunPod or a specific region appears unavailable, check the RunPod status page before redeploying.

Quick start

1. Choose a template for your workload.
2. Select the GPU, RAM and workspace storage.
3. Click Deploy Pod.
4. Monitor the startup checkpoints.
5. Wait for the final ready message.
6. Open ComfyUI and continue with the tutorial.

Choose a template

A compatible GPU is preconfigured in each RunPod template and marked with an asterisk (*). Check the hardware guide before choosing another GPU.

Workload	Use it for	Templates	Hardware guidance
Image	Image generation and editing	View image templates	Compatible GPUs
MiniMax H3	Video and native-audio generation	View MiniMax templates	Compatible GPUs
WAN	Video generation and motion transfer	View WAN templates	Compatible GPUs
LTX	Video and audio generation	View LTX templates	Compatible GPUs

Deploy the pod

The screenshot shows RunPod's New Pod deployment page with the early-access interface enabled.

1. Choose an available GPU, preferably one marked with an asterisk (*).
2. Set a suitable minimum amount of system RAM in the filters or use the default settings.
3. Choose the Volume Disk mounted at /workspace.
4. Enable volume encryption if desired.
5. Click Deploy Pod.

View the New Pod deployment page

Choose enough system RAM

• For image generation, allocate at least twice as much system RAM as GPU VRAM.
• For video generation, allocate at least two or three times as much system RAM as GPU VRAM.

For example, 60 GB of system RAM is a reasonable selection for image generation with an RTX 4090 with 24 GB of VRAM, for video generation more then 90 Gb is prefered.

View the RAM filter

Assign volume disk (legacy)

• Choose volume disk for the fastest access.
• Preferably use or increase the volume assigned by the template.
• Encrypt volume if desired.

View the Volume selection screen

Monitor startup

Open Logs or Connect after deployment. Use these checkpoints to distinguish a normal startup from an underperforming pod.

Phase	Healthy indication	When to investigate	Recommended action
Docker download starts	Activity begins within 1 minute	No download activity after 1 minute	Try another region
Docker extraction starts	Extraction begins within 3 minutes	Download continues without extraction after 3 minutes or stalls around 33%	Redeploy in another region
Container download and extraction	Completes in approximately 4–6 minutes	Progress is stalled or unusually slow	Compare with another pod or region
Copy ComfyUI to /workspace	Completes in 0–120 seconds	Stalls or exceeds 130 seconds	Compare storage performance with another pod
Model download	Sustained speed above 200 MB/s is acceptable	A model download fails	Let startup finish, then restart the pod
Startup complete	The final ready message appears	No final message after preceding steps complete	Review the container logs

Docker download and extraction

RunPod first downloads and extracts the container. This normally takes approximately 4–6 minutes, depending on the region. The extraction phase ends with a completion message in the system logs.

View a normal Docker download and extraction

Download starts

Download and extraction complete

Container startup

After extraction, the container starts and copies ComfyUI to /workspace. This copy usually takes between 0 and 110 seconds. A transfer rate of approximately 1 GB per 30 seconds is acceptable.

View normal container startup

Downloading models from the Hugging Face Hub

Model download speeds depend on both network and storage performance. A speed above 200 MB/s is acceptable. If a download fails, let the startup script finish and then restart the pod so that it can download any missing models.

View normal model downloads

Pod ready

The pod is ready when the final startup message appears in the container log.

Examples of healthy pods

These examples show pod configurations after a successful deployment. Hardware and startup times can vary by region and host.

View successful pod examples

Troubleshooting

Use this section when a startup checkpoint is missed. The normal deployment path above remains the quickest reference when startup is progressing as expected.

CUDA errors during deployment

• Try another region.
• Set the CUDA filter to CUDA 13.0 in the RunPod console.

View the CUDA filter

Container download does not start

If the image does not begin downloading within 1 minute, redeploy the pod in another region.

View an example with no download activity

Container does not begin extracting

If the download continues without extraction after 3 minutes, redeploy the pod in another region.

View a slow download without extraction

Slow copy to the workspace

The time required to copy ComfyUI to /workspace is a useful indicator of storage and system performance. If the copy stalls or exceeds 120 seconds, compare it with another pod. Slow storage can also affect model extraction and loading models into GPU memory.

Compare excellent, acceptable and poor copy performance

Excellent performance

Acceptable performance

Poor performance

Continue with the tutorial

When the final ready message appears, connect to your pod and open ComfyUI.