Hugging Face's logo
Back to Articles

Ultimate ComfyUI & SwarmUI on RunPod Tutorial with Addition RTX 5000 Series GPUs & 1-Click to Setup

Community Article Published June 10, 2025
Upvote
1
Furkan Gözükara's avatar
Furkan Gözükara
MonsterMMORPG

Tutorial Link : https://youtu.be/R02kPf9Y3_w

Ultimate ComfyUI & SwarmUI on RunPod Tutorial with Addition RTX 5000 Series GPUs & 1-Click to Setup

image

This is the ultimate tutorial for anyone looking to install and use ComfyUI or SwarmUI on the RunPod cloud platform. Whether you want to scale your AI generation or you're a "GPU poor" professional needing access to top-tier GPUs, this guide provides a step-by-step process.

🚀 Overview

This guide, based on the SECourses tutorial, will walk you through:

  1. Installing ComfyUI on RunPod step by step.
  2. Installing SwarmUI on RunPod step by step.
  3. Connecting SwarmUI to your ComfyUI backend to leverage its ultimate performance and optimizations.

The custom installers used in this tutorial come pre-packaged with essential libraries to maximize performance, including:

  • Torch 2.7
  • CUDA 12.8
  • xFormers
  • Sage Attention
  • Flash Attention
  • Accelerate
  • Triton
  • DeepSpeed
  • ComfyUI Manager
  • and more!

⬇️ Required Files & Installers

⚙️ Tutorial Steps (Video Chapters)

  • 0:00 Introduction to ComfyUI & SwarmUI Installation on RunPod
  • 0:22 Advanced Features of the One-Click Installer (Sage Attention, xFormers, Blackwell)
  • 1:03 Demonstration on RTX Pro 6000 GPU & VRAM Optimization
  • 1:52 Introducing the High-Speed Unified Model Downloader
  • 2:06 Starting the Tutorial: Downloading & Preparing the Installer Files
  • 2:45 Registering and Setting Up Your RunPod Account & Billing
  • 3:04 Selecting the Optimal RunPod GPU (RTX Pro 6000) and Server
  • 3:54 - Critical Step:- Configuring the RunPod Pod Template (PyTorch 2.2.0)
  • 4:20 Setting Volume Disk, Container Disk, and Exposing HTTP Ports
  • 5:09 Deploying the Pod and Accessing the Jupyter Lab Interface
  • 5:46 Uploading and Extracting the ComfyUI Installer on RunPod
  • 6:16 Running the ComfyUI Installation Command in the Terminal
  • 7:22 How to Start ComfyUI with Custom Launch Arguments (e.g., --gpu-only)
  • 8:13 Overview of Included Libraries (Manager, Flash Attention, DeepSpeed, Triton)
  • 9:15 Using Sage Attention vs. xFormers for Optimal Performance
  • 9:42 Connecting to the Live ComfyUI Web Interface
  • 10:27 Introducing the Unified Model Downloader for ComfyUI & SwarmUI
  • 11:01 Running the Model Downloader and Configuring the Download Path
  • 11:43 Downloading FLUX Models for a ComfyUI Image Generation Test
  • 13:22 Configuring the ComfyUI Workflow with Newly Downloaded FLUX Models
  • 15:01 First Image Generation Test: Analyzing the Incredible Speed
  • 15:53 Terminating ComfyUI and Preparing for SwarmUI Installation
  • 16:25 Installing SwarmUI and Setting it Up via the Cloudflared Link
  • 16:50 How to Configure SwarmUI to Use ComfyUI as a Self-Starting Backend
  • 17:50 Downloading Wan 2.1 Video Generation Models with the Downloader
  • 19:22 Importing Presets and Setting up a Wan 2.1 Video Generation in SwarmUI
  • 20:21 First Video Generation Test with Wan 2.1 (GGUF Version)
  • 21:26 Analyzing GGUF Performance and Deciding to Test the FP16 Model
  • 22:54 Downloading the FP16 Wan 2.1 Model for a Performance Comparison
  • 24:01 Installing Extra Features like RIFE Frame Interpolation
  • 24:41 Re-configuring SwarmUI to Use the FP16 Model with 16-bit Precision
  • 25:52 Analyzing FP16 Performance and VRAM Usage (57GB)
  • 26:28 Final Experiment: Using the Official Wan 2.1 Preset for Best Results
  • 27:28 Final Results, Conclusion, and How to Get Further Help

🧠 What is ComfyUI?

ComfyUI is an open-source, node-based interface for Stable Diffusion, enabling users to create complex generative art and image processing workflows. Its modular design allows for intuitive drag-and-drop connections, making it accessible for artists and developers. With extensive customization, plugin support, and a vibrant community, ComfyUI empowers users to experiment with AI-driven creativity efficiently.

📚 Additional Tutorials & Resources

💬 Community & Support

Community

Sign up or log in to comment

Upvote
1