ComfyUI with RunPod: Pod Deployment + Serverless API Setup

ComfyUI Course Resources

Everything you need for both pod deployment and serverless API setup

🔧 1. Create Required Accounts

• RunPod – for GPU pods and future serverless functions:
  Referral link: https://runpod.io?ref=ckrxhc11
  👉 Signing up and depositing $10 using this referral may grant you a one-time credit bonus between $5–$500.
• HuggingFace – required for accessing and downloading AI models:
  https://huggingface.co/
• Civitai – used to download community models:
  https://civitai.com/

💳 2. Fund Your RunPod Account

Ensure you have at least $10 in your RunPod account to spin up GPU pods and receive the bonus credit (if using referral).

Referral link: https://runpod.io?ref=ckrxhc11

💾 3. Create Network Storage

• Go to the "Network Volumes" section in RunPod.
• Create a new volume with at least 40 GB of space.
• This will allow persistent storage between pod sessions.

⚙️ 4. Deploy a Pod with GPU

1. From the RunPod dashboard, click "Deploy Pod".
2. Select a GPU (e.g., RTX A5000 for better performance).
3. Under Templates, choose: ComfyUI Manager – Permanent Disk – torch 2.4
4. Attach your network volume during setup.

📓 5. Launch the Environment

• Wait for the pod to fully install (may take several minutes).
• Once the pod is ready, click "Connect" > "Open in Jupyter Notebook".
• In the Jupyter interface, open the terminal and run:

🧱 6. Import or Create Your Workflow

• Use drag-and-drop to import your ComfyUI workflow.
• The ComfyUI Manager interface will load automatically after executing the run script.

📦 7. Install Missing Nodes

If required, install custom ComfyUI nodes such as:

🧠 8. Install Necessary Models

All commands below use curl and require a valid HuggingFace or Civitai token.
Replace "Your token here" with your actual token.

🔑 9. Collect RunPod Access Credentials

From the pod dashboard, obtain:

• RunPod API Key (for remote access or automation)
• Pod username/password (for Jupyter and SSH)

🧪 10. Open GitHub App Repo and Test

• Open the GitHub repository for your custom app.
• Update the .env file with your API key.
• Install dependencies using the provided README.
• Run the app and verify it functions properly.
• You can also modify this app to test your own custom workflows.

📌 The following Part 2 section covers Serverless deployment which continues this setup guide. Steps 1–8 from Part 1 remain the same. Steps 9 and 10 are replaced in Part 2.

ComfyUI Serverless Endpoint Deployment Guide

This documentation provides a complete walkthrough for setting up a serverless endpoint for ComfyUI, including model preparation, environment configuration, deployment, and testing. Two testing methods are covered: Postman and a custom web app.

✅ Prerequisites

• You have followed Part 1 of this guide exactly
• A working RunPod environment (optional)
• GitHub account with private repo access
• Hugging Face account with access token
• Custom ComfyUI nodes (if applicable)

📂 Step 1: Prepare the Model Folder

• Move your model folder into the root directory of ComfyUI.
• Delete any other unnecessary folders to reduce size and clutter.
• If you made modifications to your ComfyUI, download/export your custom API and workflows.

🛑 Step 2: Terminate the Pod (Optional)

• If not using the RunPod instance anymore, you can terminate the pod.
• All files are stored in Network Storage and can be accessed by the serverless endpoint.

⚙️ Step 3: Modify the Endpoint File

Use the official ComfyUI RunPod serverless template as a base.

🔐 Step 4: Add Hugging Face Token to Dockerfile

Add your Hugging Face token to the Dockerfile for model access during deployment.

📸 Step 5: Create Snapshot File

• Name it as snapshot_example.json.
• Include your custom nodes, e.g., ComfyUI-Inpaint-CropAndStitch.
• Add GitHub links to other custom nodes if needed.

⬆️ Step 6: Upload to Private GitHub Repository

• Create a private repo and push your ComfyUI files, Dockerfile, and snapshot.
• Make sure large model files are not tracked unless necessary.

☁️ Step 7: Deploy as Serverless Endpoint

• Connect your GitHub repo on your runpod.
• Choose the github repo.
• Attach your network drive.

🧪 Step 8: Testing the Endpoint

Two methods available:

• Method 1: Postman
• Method 2: Custom Web App

⚠️ Notes

• If you modified ComfyUI from Part 1, update your Postman and app code accordingly.
• Match API structure, input keys, and output format.

📌 Summary Checklist