NOTE: This tool is mostly written by AI.

comfypod

CLI tool for running ComfyUI on RunPod with persistent model storage.

Why?

Running ComfyUI on cloud GPUs has a few pain points:

• GPU time is expensive - Downloading models (often 10-50GB+) wastes expensive GPU hours
• Data disappears - Stopping a pod deletes everything, forcing re-downloads every session
• Setup is repetitive - Manually installing the same models and custom nodes each time
• Security concerns - RunPod proxy URLs have no built-in authentication

comfypod solves these by:

• Using cheap CPU pods for model downloads, storing them on a persistent Network Volume
• Keeping your models safe when GPU pods are stopped (pay only for storage)
• Defining your environment in a config file for reproducible setups
• Adding Basic Auth protection automatically

Features

• Network volume for persistent model storage (pay only for storage when not using GPU)
• Automatic model downloading from HuggingFace and Civitai
• Custom node installation
• Basic auth protection for web UI
• Optional idle auto-termination to avoid paying for an unused GPU pod

Installation

npm install @mtsmfm/comfypod

Setup

1. Create comfypod.config.ts:

import { defineConfig } from "@mtsmfm/comfypod";

export default defineConfig({
  dataCenterId: "EU-CZ-1",
  gpu: {
    typeIds: ["NVIDIA GeForce RTX 3090"],
  },
  models: [
    {
      url: "https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors",
      dest: "checkpoints/sd_xl_base_1.0.safetensors",
    },
  ],
  customNodes: [{ repo: "https://github.com/ltdrdata/ComfyUI-Manager" }],
});

Tokens default to environment variables (RUNPOD_API_KEY, HF_TOKEN, CIVITAI_TOKEN) but can be overridden in config.

Usage

# Estimate network volume size needed
npx @mtsmfm/comfypod estimate

# Download models to network volume (uses cheap CPU pod)
npx @mtsmfm/comfypod setup

# Start GPU pod with ComfyUI
npx @mtsmfm/comfypod start

# Check status
npx @mtsmfm/comfypod status

# Stop and delete GPU pod (keeps network volume)
npx @mtsmfm/comfypod stop

# Delete everything including network volume
npx @mtsmfm/comfypod cleanup

Options

npx @mtsmfm/comfypod <command> --config <path>
npx @mtsmfm/comfypod <command> -c <path>

Configuration

| Option | Description | Default | | ---------------------- | --------------------------------- | ------------------------------- | | dataCenterId | RunPod data center ID | (required) | | tokens.runpodApiKey | RunPod API key | $RUNPOD_API_KEY | | tokens.hfToken | HuggingFace token | $HF_TOKEN | | tokens.civitaiToken | Civitai token | $CIVITAI_TOKEN | | networkVolume.name | Network volume name | comfyui-models | | networkVolume.sizeGb | Network volume size (GB) | 50 | | cpu.podName | CPU pod name | comfyui-downloader | | cpu.image | CPU pod Docker image | node:24-slim | | cpu.flavorIds | CPU flavor IDs | ["cpu3c"] | | cpu.volumeMountPath | Volume mount path (CPU) | /workspace | | gpu.podName | GPU pod name | comfyui-gpu | | gpu.image | GPU pod Docker image | yanwk/comfyui-boot:cu128-slim | | gpu.typeIds | GPU types to use | (required) | | gpu.entrypoint | Entrypoint script | /runner-scripts/entrypoint.sh | | gpu.proxyPort | Proxy port | 80 | | gpu.targetPort | ComfyUI port | 8188 | | gpu.volumeMountPath | Volume mount path (GPU) | /root/ComfyUI/models | | gpu.customNodesDir | Custom nodes directory | /root/ComfyUI/custom_nodes | | gpu.preCommands | Commands to run before entrypoint | [] | | gpu.env | Environment variables for GPU pod | { CLI_ARGS: "--cache-lru 0" } | | gpu.idleTimeoutMin | Auto-terminate the pod after this many minutes with an idle ComfyUI queue. 0 disables. | 0 | | models | Models to download | [] | | customNodes | Custom nodes to install | [] |

Full Example

import { defineConfig } from "@mtsmfm/comfypod";

export default defineConfig({
  dataCenterId: "EU-CZ-1",

  tokens: {
    runpodApiKey: "your-runpod-api-key",
    hfToken: "your-huggingface-token",
    civitaiToken: "your-civitai-token",
  },

  networkVolume: {
    name: "my-comfyui-models",
    sizeGb: 100,
  },

  cpu: {
    podName: "my-downloader",
    image: "node:24-slim",
    flavorIds: ["cpu3c"],
    volumeMountPath: "/workspace",
  },

  gpu: {
    podName: "my-comfyui",
    image: "yanwk/comfyui-boot:cu128-slim",
    typeIds: ["NVIDIA GeForce RTX 3090"],
    entrypoint: "/runner-scripts/entrypoint.sh",
    proxyPort: 80,
    targetPort: 8188,
    volumeMountPath: "/root/ComfyUI/models",
    customNodesDir: "/root/ComfyUI/custom_nodes",
    preCommands: ["echo 'Starting ComfyUI...'"],
    env: {
      CLI_ARGS: "--cache-lru 0",
    },
    idleTimeoutMin: 30,
  },

  models: [
    {
      url: "https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors",
      dest: "checkpoints/sd_xl_base_1.0.safetensors",
      sha256:
        "31e35c80fc4829d14f90153f4c74cd59c90b779f6afe05a74cd6120b893f7e5b",
    },
  ],

  customNodes: [
    { repo: "https://github.com/ltdrdata/ComfyUI-Manager" },
    {
      repo: "https://github.com/cubiq/ComfyUI_IPAdapter_plus",
      name: "IPAdapter",
    },
  ],
});

License

MIT