Furkan Gözükara
patreon
Just published a tutorial that shows how to properly install ComfyUI, SwarmUI, use installed ComfyUI as a backend in SwarmUI with absolutely maximum best performance such as out of the box Sage Attention, Flash Attention, RTX 5000 Series support and more. Also how to upscale images with max quality
🕑 Added 2025-05-04 20:50:25 +0000 UTCTutorial Link
Tutorial Information
If you want to generate the very best AI videos and images on your Windows computer locally this is the tutorial that you were looking for. Literally 1-click to install most powerful and advanced generative AI interface SwarmUI (with Flash Attention, Sage Attention, Triton, DeepSpeed, xFormers, RTX 5000 series perfect compatibility) and download the very best AI image and video generation models with ultra advanced model downloader Gradio app. SwarmUI utilizes the famous and most powerful, advanced, performant and optimized ComfyUI as a backend. So SwarmUI is the ultimate generative AI tool at the moment with vast amount of features and constant updates.
Tutorial Important Download Links App Links
🔗Follow below link to download the zip file that contains SwarmUI installer and AI models downloader Gradio App - the one used in the tutorial ⤵️
🔗Follow below link to download the zip file that contains ComfyUI 1-click installer that has all the Flash Attention, Sage Attention, xFormers, Triton, DeepSpeed, RTX 5000 series support ⤵️
🔗 Python, Git, CUDA, C++, FFMPEG, MSVC installation tutorial - needed for ComfyUI ⤵️
🔗 SECourses Official Discord 10500+ Members ⤵️
🔗 Stable Diffusion, FLUX, Generative AI Tutorials and Resources GitHub ⤵️
🔗 SECourses Official Reddit - Stay Subscribed To Learn All The News and More ⤵️
Tutorial Details
This tutorial will show you how to install ComfyUI properly into a VENV with 1-click our installers, then install SwarmUI instantly and then how to give propoerly installed ComfyUI as a backend to the SwarmUI and generate AI images and videos with maximum speed and quality. Moreover, this tutorial will introduce you ultimate AI model downloader Gradio app being developed by us.
Tutorial Chapters
0:00 Introduction: Fastest Local Image & Video Generation with SwarmUI on Windows
0:38 Leveraging ComfyUI Performance & Sage Attention Speed Boost in SwarmUI
1:07 Introducing the Custom Unified Model Downloader Application
1:50 Overview of Supported Models (SD, Video, Text Encoders, GGUF, VAE etc.)
2:50 Why Manually Installed ComfyUI Backend is Better for SwarmUI Performance
3:07 Live Demo Preview: High Dream GGUF Image Generation on Dual GPUs (RTX 5090 + 3090 Ti)
3:52 Tutorial Start: Overview of Installers (SwarmUI, Model Downloader, ComfyUI)
4:20 Step 1: Downloading the SwarmUI Model Downloader Installer Zip
4:33 Step 2: Downloading the Manual ComfyUI Backend Installer
4:52 Step 3: Extracting ComfyUI Installer for Backend Setup
5:05 Step 4: Running ComfyUI Backend Installation (windows_install.bat)
5:33 Configuring ComfyUI Install: Python Version & Attention Mechanisms (Flash/Sage)
6:18 Step 5: Starting SwarmUI Frontend Installation (Parallel to ComfyUI)
6:40 Configuring SwarmUI Installation Options (Theme, Custom Settings)
6:54 CRUCIAL STEP: Selecting "None" for ComfyUI Backend during SwarmUI Setup
7:21 Step 6: Adding the Manually Installed ComfyUI Backend in SwarmUI Settings
7:43 Configuring Backend Path (main.py) & Enabling Sage Attention (--use-sage-attention)
8:28 Monitoring Backend Dependency Installation & VENV Explanation
9:17 Step 7: Adding a Second ComfyUI Backend for Multi-GPU Usage (GPU ID 1)
10:00 Step 8: Launching & Setting Up the Unified Model Downloader App
10:50 Using the Model Downloader: Downloading Flux Models Bundle Example
11:14 Understanding Downloader Speed (HF Transfer) & Progress Monitoring
11:50 Downloader Features: Queue System & No Re-Downloads (Wan 2.1 Example)
12:20 Accessing 'How-to-Use' Links for Models within the Downloader
13:22 Step 9: Starting SwarmUI & Refreshing Models After Downloading
14:08 First-Time GGUF Model Load: Installing GGUF Support Extension in SwarmUI
14:32 Manually Setting Model Metadata (Architecture) if Needed (HiDream Example)
14:48 Detailed GGUF Support Installation Walkthrough & Backend Restart
15:51 Ignoring C Link Fatal Error on Windows
16:23 Usage Demo: Generating Image with Custom Flux DreamBooth Model
16:41 Configuring Generation Parameters (Prompt, Steps, Sampler, CFG, Guidance)
17:07 Initiating Generation & Monitoring Model Loading on GPUs
17:50 Troubleshooting Face Inpainting & Downloading YOLO Models via Downloader
18:31 Advanced Feature Demo: Using the Refine Upscale Option
18:43 Configuring Upscale Settings (Refiner %, Upscaler Model)
19:19 Running Generation with Upscaling & Performance Considerations
19:40 Comparing Results: Original vs. High-Quality Upscaled Image Detail
19:52 Fine-tuning Face Inpainting: Adjusting Mask Parameters (Blur/Grow)
20:23 Conclusion & Where to Find Links and Ask Questions
Comments
Furkan Gözükara
yes you have to install GGUF it is separate library. so sorry for late reply
Tenally
I downloaded the Hidream bundle. When I load it, I do not get the popup to install GGUF Support. Does the bundle download what is required where separate popup for GGUF Support is not necessary?
Furkan Gözükara
yes it should work. you can leave the rest as in tutorial. cfg is 1. it should work with lora
Hockey
Settings for HiDream_I1_Dev_GGUF_Q6_K? Steps, Scheduler, Sampler, CFG and DCF? Does it work with LORA?
Furkan Gözükara
yes install comfyui with our installer on runpod : https://www.patreon.com/posts/105023709 then install swarmui with our installer and give its backend : https://www.patreon.com/posts/114517862 our comfyui installer has support for rtx 5090 but pick the template : RunPod Pytorch 2.2.0 runpod/pytorch:2.2.0-py3.10-cuda12.1.1-devel-ubuntu22.04
Marcin Jarosz
Hey Furkan do you have a way to properly install and run SwarmUi in RunPod so that RTX5090 support works, currently SwarmUI does not work on RTX5090, the second issue is running ComfyUI backend installed by myself, there is no support for I prefer to use segment instead of Yolov because Yolov only corrects the face, in segment I have individual body parts
