Skip to content

Deploy on RunPod

One-Click Deploy

Use the public template:

Deploy on RunPod

This opens the RunPod GPU selection page with the ComfyUI Studio template pre-configured.

GPU Selection

Any NVIDIA GPU works. Choose based on your use case:

Use Case Recommended VRAM Why
WAN 2.2 video (14B models) A100 80GB, H100, B200 80+ GB 14B models need ~28 GB VRAM at FP16, more with long videos
HunyuanVideo / CogVideoX A100 40GB, L40S 40+ GB Large diffusion models
Flux image generation RTX 4090, A6000, L40S 24-48 GB Flux dev FP16 needs ~24 GB
SDXL / SD 1.5 images RTX 3090, A5000, L4 16-24 GB Smaller models, fast generation
Budget / experimentation RTX 3060, T4 12-16 GB Works with FP8/GGUF quantized models

Environment Variables

Set these in the RunPod template before launching:

Variable Required Value
API_KEY Yes A strong password for the web UI. Default is changeme — change it.
CIVITAI_API_KEY Recommended Your CivitAI API token. Get it from civitai.com/user/account → API Keys. Without this, CivitAI model downloads and metadata fetch won't work.
HF_TOKEN Recommended Your HuggingFace access token. Get it from huggingface.co/settings/tokens. Without this, gated models (Flux, WAN, Hunyuan) can't be downloaded.

Disk Configuration

Setting Recommended Why
Container Disk 5 GB The Docker image is managed separately — 5 GB is enough for logs and temp files
Network Volume 50+ GB AI models are large. A single WAN 2.2 FP16 model is ~26 GB. Start with 50-100 GB and expand as needed.

Ports

The template exposes:

Port Service Access
8000 ComfyUI Studio https://<POD_ID>-8000.proxy.runpod.net — password-protected web UI
8188 ComfyUI https://<POD_ID>-8188.proxy.runpod.net — graph editor, no authentication

ComfyUI has no authentication

Port 8188 (ComfyUI graph editor) has no password protection. Anyone who knows the URL can access it. Consider not exposing this port if you don't need the graph editor. Advanced users can access it via SSH tunnel instead.

First Boot

The first boot takes 2-5 minutes:

  1. Bootstrap clones the application from GitHub
  2. ComfyUI is copied from the Docker image to the persistent volume
  3. Both services start

Watch the boot progress in RunPod → Pod → Logs. Look for === Services started ===.

Accessing the Web UI

Open https://<POD_ID>-8000.proxy.runpod.net and enter your API_KEY at the login page.

Subsequent Boots

After the first boot, subsequent restarts are faster (~30 seconds) because:

  • Bootstrap is skipped (application already on the volume)
  • ComfyUI copy is skipped (already on the volume)
  • Only the two services need to start

Updating the Application

Click Check for Updates on the Home page. The backend fetches the latest code from GitHub and updates changed components. If the backend code changed, it restarts automatically (~2 seconds).

No need to restart the pod or pull a new Docker image for application updates. Docker image rebuilds are only needed when the infrastructure changes (new custom nodes, new Python packages, CUDA version change).

SSH Access

RunPod pods support SSH for advanced access:

ssh <POD_ID>@ssh.runpod.io -i ~/.ssh/id_ed25519

From inside the pod you can:

  • View logs: tail -f /var/log/comfyui.log or /var/log/admin.log
  • Access ComfyUI on localhost: curl http://localhost:8188/system_stats
  • Browse files: ls /workspace/studio/
  • Compile llama-server manually if not in the Docker image