Feel the Power: Run ComfyUI on Cloud GPUs - Full VM Setup Guide

By Heiko Polinski•October 15, 2025

TutorialsComfyUIGPU RentalDocker

Running ComfyUI locally is great, but unless you're running a RTX 5090 or similar high end hardware, you will run into bottlenecks sooner than later.

So using cloud hosted GPUs from platforms like CloudRift will not only save your hardware, it's also give you much faster results. Plus you only pay for what you use. In a previous article I gave you a walkthough on how to set up ComfyUI in a pre-configured container. Which is great to start out, but as soon as you want to add checkpoints and configure your workflows more, you'll need some additional freedom.

If that's what you're looking for, and you want to configure your own environment instead of using a pre-built container, this guide shows how to launch ComfyUI on a rented GPU-powered VM in minutes.

This is Part 2 of our ComfyUI series. Part 1: Preconfigured Containers →

Prerequisites

Before starting, you'll need a GPU VM:

1. Register at CloudRift Console
2. Add a few dollars in credits
3. Click "New" and select "VM" under the Virtualization step

1. Choose Ubuntu 24.04 as your OS

Note: Port Forwarding Setup

Some of our datacenter providers use shared public IPs with port forwarding. In this case, during the Software step of the rental process, you'll see a Port Mapping section. Make sure to add port 8188 there — this is the port ComfyUI uses, and it needs to be exposed for you to access the UI in your browser.

Once your VM is running, continue below.

Quick Overview

• Works on Ubuntu 22.04 and 24.04
• Tested on RTX 4090 instances
• No extra dependencies
• Takes less than 3 minutes

Step-by-Step Setup

1. Connect to your VM

To connect to your VM via SSH, run the following command.

Open your terminal (or command prompt on Windows) and run:

ssh riftuser@<VM_IP>

Replace <VM_IP> with your actual VM's IP address. You'll find it in your Dashboard.

Troubleshooting: Host Key Verification Failed

Sometimes when connecting, your terminal may throw a "Host Key Verification Failed" error with a message like "Someone could be eavesdropping on you right now (man-in-the-middle attack)!" or "It is also possible that a host key has just been changed." This typically happens when a VM's IP address has been reassigned or the server has been re-provisioned.

Step 1: Remove the old host key from your known_hosts file:

ssh-keygen -R "[<IP>]:<port>"

Replace <IP> and <port> with your VM's IP address and SSH port. For example: ssh-keygen -R "[211.21.50.85]:57019"

If your VM uses a standard SSH port (22) without port forwarding, you can simply use: ssh-keygen -R <IP>

Step 2: Try to connect to your VM again:

ssh riftuser@<VM_IP>

When you run this command, you should see a warning message similar to the following:

The authenticity of host '<hostname> (<IP address>)' can't be established.
RSA key fingerprint is <fingerprint>.
Are you sure you want to continue connecting (yes/no)?

Type yes and press Enter to continue. You should now be connected to the remote server, and the host key for that server will be added to your known_hosts file.

Source: linuxblog.io

2. Verify GPU

Once connected to your VM via SSH, verify your GPU is accessible by running:

nvidia-smi

If you see your GPU listed, everything is ready.

3. Check if Docker is installed

docker --version

This Ubuntu Version should come with Docker pre-installed. If missing you can run:

sudo apt update && sudo apt install -y docker.io
sudo systemctl enable docker --now

4. Fix permissions

sudo usermod -aG docker $USER
newgrp docker
docker ps

An empty list confirms Docker is working without sudo.

5. Confirm GPU access (optional)

For Ubuntu 22.04:

docker run --rm --gpus all nvidia/cuda:12.3.2-base-ubuntu22.04 nvidia-smi

For Ubuntu 24.04:

docker run --rm --gpus all nvidia/cuda:12.4.0-base-ubuntu24.04 nvidia-smi

6. Launch ComfyUI

mkdir -p ~/comfyui/models/checkpoints

docker run -d \
  --name comfyui \
  --gpus all \
  -p 8188:8188 \
  -e WEB_ENABLE_AUTH=false \
  -v ~/comfyui/models:/opt/ComfyUI/models \
  ghcr.io/ai-dock/comfyui:latest-cuda

7. Add a model checkpoint

cd ~/comfyui/models/checkpoints
wget -O flux1-schnell-fp8.safetensors \
  "https://huggingface.co/Comfy-Org/flux1-schnell/resolve/main/flux1-schnell-fp8.safetensors"
docker restart comfyui

This is the Flux Schnell model, which is great to start with. Obviously you get to choose whichever model you like.

8. Open ComfyUI in your browser

http://<VM_IP>:8188

Replace <VM_IP> with your VM's IP address.

Note: Port Forwarded VMs

If your provider uses port forwarding (shared public IP), you won't use port 8188 directly. Instead, go to your CloudRift dashboard, click "View all mappings" on your instance, and find the external port mapped to 8188. Then open ComfyUI at http://<IP>:<mapped_port>/ — for example: http://211.21.50.85:57018/

If the port is blocked, go back to your terminal and run:

sudo ufw allow 8188/tcp

Once ComfyUI loads, you'll see the default workflow. You can adjust your prompt, set the CFG to 1 for the Flux Schnell model for best results, and click "Queue Prompt" to start generating images.

9. Optional management

Stop the container:

docker stop comfyui

Start it again:

docker start comfyui

Remove the container completely:

docker rm comfyui

Result

ComfyUI runs on your GPU VM with the Flux Schnell FP8 checkpoint loaded.

Add more models to ~/comfyui/models/checkpoints or build your workflow directly in the browser.

TL;DR

Question: How do I install and run ComfyUI on a GPU VM on CloudRift?

Answer: Connect to your Ubuntu 22.04 or 24.04 VM, verify the GPU with nvidia-smi, install Docker, fix permissions using sudo usermod -aG docker $USER, then run the ComfyUI Docker image and add your model checkpoint. Access it at http://<VM_IP>:8188.

Video Walkthrough

What's Next?

Explore more ways to rent a GPU for ComfyUI:

• Quick Setup: Rent a GPU for ComfyUI in Under 2 Minutes → - Pre-configured container (fastest option)
• Complete Setup Guide: How to Rent a GPU for ComfyUI → - In-depth tutorial with template and CLI methods

For more advanced workflows, custom model management, and persistent storage configurations, look out for future articles!