How to Install and Use ComfyUI and SwarmUI on Massed Compute and RunPod Private Cloud GPU Services

How to Install and Use ComfyUI and SwarmUI on Massed Compute and RunPod Private Cloud GPU Services

Full tutorial link > https://www.youtube.com/watch?v=bBxgtVD3ek4

If your GPU is not strong enough to run Generative AI models this is the tutorial that you need. Or you want to scale your generation speed by using multiple GPUs at the same time again this is excellent tutorial. In this tutorial I will show how to setup ComfyUI and SwarmUI literally 1-click on Massed Compute and RunPod and use your most liked best image and video generation models like Qwen, FLUX, Wan 2.2 and more.

🔗 Important Links:

Download ComfyUI Installer: https://www.patreon.com/posts/105023709

Download SwarmUI Installer and Model Downloader: https://www.patreon.com/posts/114517862

Previous Detailed Windows Tutorial (Recommended Watch): https://youtu.be/c3gEoAyL2IE

⏰ TIMESTAMPS / CHAPTERS

• 00:00:00 Introduction & Tutorial Goals

• 00:00:39 Downloading the ComfyUI Installer & Reading Update News

• 00:01:06 Downloading the SwarmUI Installer & Checking Changelogs

• 00:01:25 Extracting ComfyUI & Opening the Massed Compute Instructions

• 00:01:48 Deploying on Massed Compute: GPU Selection

• 00:02:03 Applying the 'SECourses' Coupon Code

• 00:02:30 Choosing a Multi-GPU Machine for the Demo

• 00:02:55 Installing the ThinLinc Remote Desktop Client

• 00:03:12 Crucial: Configuring ThinLinc Local Devices & Shared Drives

• 00:03:52 Connecting to the Massed Compute Desktop

• 00:04:13 Transferring Installer Files to the Remote Machine

• 00:05:18 Installing ComfyUI via Terminal Command

• 00:06:31 Updating the Pre-Installed SwarmUI on the Machine

• 00:07:00 Preparing the SwarmUI Model Downloader

• 00:07:48 Launching the Downloader & Downloading Model Bundles

• 00:08:46 Launching SwarmUI with a Public Cloudflare Link

• 00:09:11 Configuring the First Backend (GPU 0) with Sage-Attention

• 00:10:04 Adding and Configuring the Second Backend (GPU 1)

• 00:10:33 Importing the 'Amazing Swarm' Presets

• 00:11:04 Live Demo: Generating Realistic Images

• 00:12:48 Monitoring the Multi-GPU Generation Process

• 00:13:32 How to Download Your Generated Images (Two Methods)

• 00:14:12 IMPORTANT: How to Stop Billing by Deleting the Machine

• 00:15:07 Part 2: Starting the RunPod Installation

• 00:15:26 Deploying a RunPod Pod: Choosing the Right Template

• 00:16:24 Setting Pod Volume Size and Overrides

• 00:16:41 Troubleshooting: Handling a Pod That Won't Start

• 00:18:07 Uploading & Installing ComfyUI on RunPod

• 00:19:30 Uploading & Installing SwarmUI on RunPod

• 00:20:34 First-Time SwarmUI Setup Wizard (Important Settings)

• 00:21:04 Configuring Multi-GPU Backends on RunPod

• 00:22:14 Downloading Models Using the SwarmUI Downloader on RunPod

• 00:23:51 Importing Presets into SwarmUI on RunPod

• 00:24:31 Live Demo: Generating Images on RTX 4090s

• 00:25:52 Downloading Your Images from the RunPod Workspace

• 00:26:37 RunPod Billing: Stopping vs. Terminating Your Pod

• 00:27:16 Conclusion & Final Thoughts

🚀 Unleash the full power of AI image and video generation on the cloud! This comprehensive tutorial is your step-by-step guide to installing and configuring SwarmUI and ComfyUI on two of the most popular cloud GPU platforms: Massed Compute and RunPod.

Learn how to set up a powerful multi-GPU workflow to generate stunning, ultra-realistic images and videos at incredible speeds. We'll cover everything from deploying your first machine to downloading models, importing our exclusive presets, and running your first generations. Most importantly, we'll show you how to manage your instances to save money!

Whether you're new to cloud computing or looking to scale up your AI art projects, this guide has you covered.

💻 In this tutorial, you will learn how to:

Part 1: Massed Compute

Deploy a high-performance, multi-GPU machine.

Connect to your remote desktop using the ThinLinc client.

Install ComfyUI and the pre-installed SwarmUI from scratch.

Use the SwarmUI Model Downloader to get all the necessary models and bundles.

Configure SwarmUI backends to utilize multiple GPUs simultaneously for maximum speed.

Generate images and access your files from your local computer.

Properly terminate your machine to stop billing.

Part 2: RunPod

Deploy a multi-GPU pod using the correct PyTorch template.

Troubleshoot common connection issues.

Install both ComfyUI and SwarmUI in your RunPod workspace.

Set up SwarmUI backends for parallel processing on multiple GPUs.

Download models and presets for immediate use.

Understand the difference between stopping and terminating a pod to manage costs effectively.

Video Transcription

• 00:00:00 Greetings everyone. Today I am going to show  you how to install SwarmUI and ComfyUI on  

• 00:00:06 Massed Compute and RunPod and use all the  presets we have, generate amazing images  

• 00:00:12 or videos. If you remember, I had shown how to  generate ultra-realistic images with SwarmUI by  

• 00:00:20 using the Wan 2.2. We had published the  tutorial for Windows like two days ago,  

• 00:00:26 so please watch this tutorial to learn how to use  it in details. Today I will show how to install  

• 00:00:33 and use it on Massed Compute and RunPod,  but it won't be as detailed as this one.

• 00:00:39 So first of all, let's download the ComfyUI  installer. It is version 57 right now. You see  

• 00:00:45 the latest zip file is here. Also, I recommend  you to scroll down and read the news. You see,  

• 00:00:52 I have compiled a new stage attention and it  has fixed black outputs on RTX 3000 series.  

• 00:01:00 Maybe more of the black outputs are fixed, I  don't know yet, but this is an improvement.

• 00:01:06 Then download the SwarmUI installer. It  is version 89 right now. We are updating,  

• 00:01:13 adding more features. You can read all  the news by scrolling down and see the  

• 00:01:19 changelogs. Then move the zip files into any  folder. So I moved them here. First of all,  

• 00:01:25 I will extract the ComfyUI.  Extract all, like this.

• 00:01:30 Then enter inside the folder and you will see that  there is Massed Compute instructions .txt file.  

• 00:01:35 Double click it. We have some links and tutorials  here if you want to watch them later. Please use  

• 00:01:40 this link to register. After registration, go to  billing, set up some balance, then go to deploy.

• 00:01:48 In here, select your GPU. RTX Pro 6000 Blackwall  GPU is the best, but L40S is also really good.  

• 00:01:56 Let me show you their prices. So select creator,  select the SECourses, apply direct coupon like  

• 00:02:03 SECourses, verify. So RTX Pro 6000, 96 gigabyte  GPU is 1.47 dollars per hour. Let's see the  

• 00:02:12 L40S. Currently there is no L40S available, so  I will use this one. Or there is only one GPU,  

• 00:02:19 so let's select another one. For example,  let's get H100. Okay, it is also a single  

• 00:02:25 GPU left. So let's get this one. This is also  a fast one. Okay, it is also single. Wow. Okay,  

• 00:02:30 this one has two GPUs. Why I am choosing  two GPUs? So I can show you how to utilize  

• 00:02:35 both of the GPUs at the same time. Okay, click  verify. You see 1.80 dollars for 2x A100. Deploy.

• 00:02:43 So this is important, we have selected image  SECourses. Okay, now we need to wait for our  

• 00:02:49 machine to initialize. Meanwhile, if you have  not previously installed ThinLinc client,  

• 00:02:55 go to ThinLinc client. You see from  details, select the Windows installer,  

• 00:03:00 or if you are using Mac or Linux select  them. Open the installer, it will ask you,  

• 00:03:05 click yes. Click next, accept, next, install.  That's all default. Run ThinLinc client.

• 00:03:12 Important thing here is click  options, go to local devices,  

• 00:03:16 select clipboard synchronization and drives.  Then click details. Then we are going to add  

• 00:03:21 a folder. So select your folder from your  PC, any folder you wish. This is my folder,  

• 00:03:27 let me show you. You see, this is my folder.  Then make the permission read and write to not  

• 00:03:33 have any issues. Okay. Then okay. Okay,  now we just wait for initialization.

• 00:03:39 Okay, the initialization of the machine has  been completed. We are updating our image to  

• 00:03:45 become available faster. So hopefully the machine  initialization will become much faster. Then copy  

• 00:03:52 this login URL and copy your password and copy  your username. This is how we connect to our  

• 00:03:59 machine. Click connect, continue. Wait for screen  to start. Yes, it is starting. Then click start.  

• 00:04:06 So this will connect to the remote machine, the  Massed Compute desktop interface. This is running  

• 00:04:13 on remote. I will copy my files into shared  folder. You see ComfyUI and Swarm Downloader,  

• 00:04:21 I will copy them. This is my shared  folder. I will paste them here.

• 00:04:26 Then inside the ThinLinc client, it may take  a while to load depending on your location,  

• 00:04:32 your internet connection. Currently it is slow  at me. Go to home, then scroll down and go to  

• 00:04:38 ThinDrives. This is your shared folder with  your computer. Whatever you put inside here,  

• 00:04:45 whether on your PC or in Massed Compute, it  will be synchronized. So you can download  

• 00:04:50 your generated images and videos from here as  well. So ComfyUI version 57 and SwarmUI model  

• 00:04:57 downloader. Copy both of them or you can drag  and drop into downloads. Or right click here  

• 00:05:02 and paste. So you can use either drag and drop  or copy and paste. Wait for synchronization to  

• 00:05:08 be completed. You see, it shows that there is  a synchronization here, copying files. You need  

• 00:05:13 to wait this. Do not run anything inside  shared folder, always copy into downloads.

• 00:05:18 So the ComfyUI has been copied. Right click and  extract here. I will first install the ComfyUI.  

• 00:05:24 You see there is Massed Compute instructions,  double click it. Then select this install command,  

• 00:05:29 right click, copy. Go back to folder, click  this three dots icon. All of my installations  

• 00:05:36 are same. Open in terminal. All of my applications  installed same way. Right click and paste and hit  

• 00:05:43 enter. This will start the ComfyUI installation.  Now just wait it to be completed. This should be  

• 00:05:48 fairly fast on Massed Compute. We are also  working update of the image. As I said,  

• 00:05:54 we will have the latest ComfyUI, latest SwarmUI  and latest of everything. So the installations  

• 00:05:59 will get faster, but it is already really  fast. When you see the software updater,  

• 00:06:04 just click cancel. You don't need to install  anything. Just click cancel and follow the  

• 00:06:09 CMD terminal from here. You see when  I click it, it will show the terminal.

• 00:06:13 Okay, so the ComfyUI installation has been  completed. Now we can move to the next  

• 00:06:17 step, SwarmUI installation. However, it is not  needed in our machine. So the ComfyUI installation  

• 00:06:25 has been completed. Now it is time for SwarmUI,  but SwarmUI has been installed already. So control  

• 00:06:31 alt D, it will minimize everything. Then you  see there is run SwarmUI update. Double click  

• 00:06:38 this. This will update the SwarmUI to the latest  version and start it. Just wait for it to start  

• 00:06:44 at the beginning. So it is doing the update.  It is started. Now you can close this. Okay,  

• 00:06:49 it is already opened. I will just close it. I  will first download models, then I will start  

• 00:06:54 again. I will show you in a moment. So go back  to downloads. Right click and extract the model  

• 00:07:00 downloader. Enter inside the model downloader  folder. You will see that Massed Compute model  

• 00:07:06 download instructions .txt file. Copy this,  right click and copy. Go back to your folder,  

• 00:07:11 click this three dots, open in terminal and  paste it. This will start model downloader  

• 00:07:17 because we need to download models. If you put  the models inside the Massed Compute image,  

• 00:07:23 it takes too long to start. Therefore, we are  removing all the models, but we will keep the  

• 00:07:29 SwarmUI installed and few other applications  installed. So the ComfyUI installation or update  

• 00:07:34 will be faster. SwarmUI is already installed, as  you have seen, it was just immediately updated.  

• 00:07:41 But we are working on to make this machine start  faster and more lightweight. So the application  

• 00:07:48 is starting. In a moment, yes, started.  It will be automatically opened like this.

• 00:07:53 So I am going to download SwarmUI bundles and  image bundle. Let's see the download speed. It  

• 00:07:59 should be blazing fast. Okay, 200 megabytes per  second. Okay, around 300 megabytes per second,  

• 00:08:07 very decent. You can download any model  as well or all the bundles from here.  

• 00:08:13 You can type the name like one, it will list  you, or you can use URL downloader to download  

• 00:08:18 from Civitai or anywhere you want. But it is  downloading right now. If you want to learn more,  

• 00:08:24 please watch this tutorial because I have  explained it in this base Windows tutorial.  

• 00:08:29 I am not going to repeat everything here again.  So let's just wait for downloads to be completed.

• 00:08:35 Okay, once the model downloads have been  completed, let's return back to desktop.  

• 00:08:40 Control alt D, it will minimize everything.  Then I am going to start the SwarmUI with  

• 00:08:46 this one. You see run Cloudflare SwarmUI. Okay,  let's just run it. It will start SwarmUI with a  

• 00:08:53 Cloudflare link. Okay, it is started. You see  here, open link. And copy this link like this  

• 00:08:59 and open it in your browser so that it will  work really fast. It will be running on the  

• 00:09:06 Massed Compute machine, but I will be able  to use it from my computer. So first of all,  

• 00:09:11 we will add new backends. You see it is using  our existing backend. We will update it later,  

• 00:09:17 but using the installed ComfyUI is the best. So  it is here inside ComfyUI. So click control L,  

• 00:09:25 select it like this, control C, and paste it  here. So click this edit icon, paste here,  

• 00:09:31 and type main.py like this. Let me zoom out.  Okay, I cannot zoom out. Maybe from here,  

• 00:09:37 yes. You see this is the path wherever I have  installed, it is ComfyUI installation and at  

• 00:09:43 the end main.py. I am also going to add some extra  arguments, --use-sage-attention. If you get black  

• 00:09:52 outputs with your used model or preset, you can  remove this. And I will make the OverQueue 0.

• 00:09:59 So this is for the first GPU. You can see  that how many GPUs you have. We have two  

• 00:10:04 GPUs right now. So let's go back to backends  and add another Comfy self-starting. Copy this  

• 00:10:10 path again and paste here. Copy the extra  arguments. Make the GPU ID 1, make the  

• 00:10:16 OverQueue 0 and save. So now when I generate  multiple images or videos at the same time,  

• 00:10:21 it will be distributed on each GPU. Go back to  generate. Go to models and refresh. Yes, all the  

• 00:10:28 models are here. Go to presets. We don't have  the presets yet. However, we have preset here,  

• 00:10:33 so I will right click and extract this into my  own folder. So enter inside the folder and then  

• 00:10:41 import preset, choose file. Go back wherever you  have extracted it, select the preset and overwrite  

• 00:10:50 existing ones if you want to overwrite, import.  It is imported. Click refresh, sort by name.

• 00:10:57 So the rest is exactly same as in Windows  tutorial, but I will make a demo for you and let's  

• 00:11:04 also see the speed. First, we have to wait for  backends to be loaded. When we go to the logs and  

• 00:11:09 debug, we will see that it is loading. We should  just wait. If it doesn't load for a long time,  

• 00:11:15 you can just terminate all the terminals, quit  all three terminals, then run the Cloudflare  

• 00:11:21 stable SwarmUI again. It should fix the issue,  most likely case. So let's see. Sometimes it may  

• 00:11:29 take a while for Cloudflare URL to start. Yes,  it is taking some time. You can also use from  

• 00:11:36 the local URL it has from here. You see local  URL. Let's go to logs. Okay. Do we have any  

• 00:11:43 issues anywhere? Yeah, the backends loaded. Now  we need to wait for Cloudflare. For some reason,  

• 00:11:50 it didn't start. So if it doesn't start, you can  restart again. Yeah, the Cloudflare didn't start,  

• 00:11:56 so I will restart again. This can happen,  unfortunately, so you need to try. Okay,  

• 00:12:01 it gave us a new link. Okay, this one  started. So let's open this in our browser.  

• 00:12:07 Now the backend will load much quicker.  Sometimes it gets stuck, so just restart.

• 00:12:13 Yes, it is started very quickly. Presets are here.  As I said, the rest is exactly as in the Windows  

• 00:12:19 tutorial. So let's make a demo. Quick tools, reset  params, and realistic images. Cinematic image of a  

• 00:12:29 fast car. Then you can choose your aspect ratio or  resolution and generate. So the first generation  

• 00:12:36 will be done on the first model. Generate again.  The second one will be in the second model. Let's  

• 00:12:41 generate 10 images. So it will generate 10 images.  The generations will happen on both of the GPUs.  

• 00:12:48 First of all, it is loading both of the GPUs.  So let's go to logs to see. After load has been  

• 00:12:54 completed, it will be much faster. Okay, the first  generation started. Now let's just see them here.  

• 00:13:02 Yes, you see both of the GPUs are generating  images at the same time. So this way you can scale  

• 00:13:08 your generation speed. You can rent as many as  GPUs having machines, like eight GPUs. You can add  

• 00:13:15 eight backends and generate eight images at the  same time. Modern GPUs are much faster, like RTX  

• 00:13:22 6000 Pro or RTX 4090, but A100 will be also fast.  You see like 3 second IT. And we got the images.

• 00:13:32 So you can from more download, download them to  your computer, or you can go to your ThinLinc  

• 00:13:38 client, go to home, go to apps, go to stable  SwarmUI. This is where it is installed. Go to  

• 00:13:44 output and go to local. So these are where the  images will be generated. So you can copy this  

• 00:13:51 local, go to your ThinDrives, enter inside your  shared folder, paste it. So once this is pasted  

• 00:13:59 here, when you go back to your shared drive  on your PC, you will see the generated images  

• 00:14:05 are also copied here like this. This is another  way of copying, mass copying. Either way works.  

• 00:14:12 So once you are done with your generations,  you need to turn off your machine. The other  

• 00:14:16 presets are same. You can just quick tools,  reset params to default and select them,  

• 00:14:21 but you need to download their models. And  downloading their models are easy. From the  

• 00:14:26 model downloader, we have bundles for everything.  So delete the research if you have and check out  

• 00:14:31 the bundles. Image generation bundle, Qwen image  core bundle, Wan 2.2 core bundle, Wan 2.1 core  

• 00:14:38 bundle, so flux models core bundle. You can  download them and use other presets as well.

• 00:14:44 And now I will turn off my machine to not spend  any time. So click delete and delete. Once you  

• 00:14:51 delete, it will be all gone. If you stop it,  it will not stop your billing. So you need to  

• 00:14:56 delete this. And it is deleted. That's it. Now  I will start the RunPod part, so you can also  

• 00:15:02 watch that part to learn how to use on RunPod  as well. Thank you so much. Let's continue.

• 00:15:07 Open the RunPod instructions .txt file. You can  watch the links here. You can read it. I recommend  

• 00:15:13 that. Please register the RunPod from here. After  registration, sign in if it doesn't automatically  

• 00:15:19 sign in you. Then go to billing and add some  credits to your account. Then go to pods. Then  

• 00:15:26 you see there is options here. Click this deploy.  I recommend to use secure cloud and in here,  

• 00:15:34 I recommend to click additional filters and NVMe  disk and make this 100 gigabytes to get a decent  

• 00:15:42 pod. Most of the times the pods of the RunPod can  be broken. I will show with 2x RTX 4090. Select  

• 00:15:50 2x, so I can show you multiple backends and how  to generate multiple images or videos at the same  

• 00:15:56 time. Now this is confusing so many people. We  are not going to use this template. Why? Because  

• 00:16:03 in the RunPod instructions .txt file, it tells you  to use this template. So always follow the RunPod  

• 00:16:09 instructions .txt files for my applications. So  click change template, select PyTorch 2.2. Okay,  

• 00:16:17 2 GPUs edit because we are going to increase the  volume disk size to like minimum 200 gigabytes.  

• 00:16:24 If you are going to download more models, you  need more and set overrides. Then click deploy on  

• 00:16:28 demand. Just wait pod to be ready. This should be  fairly fast because we are using official PyTorch  

• 00:16:35 2.2 template. Yes. It takes few seconds. Then  click Jupyter Lab. Okay, it is not ready yet,  

• 00:16:41 so you need to try again and again until it  becomes ready. If it doesn't become ready  

• 00:16:46 like in one or two minutes, delete the pod and  get a new one. Unfortunately, because RunPod is  

• 00:16:51 extremely unpredictable. Okay, it didn't start.  I am trying again. Yes, it still didn't start.  

• 00:16:58 Try again. Okay, it doesn't start, so I will get  a new pod meanwhile to increase my chances. Okay,  

• 00:17:05 filters are remaining. Let's make this 2x, change  template, 2.2 edit, 200 gigabytes, set overrides,  

• 00:17:14 and deploy on demand. So whichever starts first,  I will delete the other one. Okay, try again. No,  

• 00:17:21 I cannot access yet. In these machines, I am doing  Qwen image full fine tuning. I am researching it  

• 00:17:29 right now. Okay, this one also failed. So I  don't need this one at the moment. The test  

• 00:17:34 has been completed. So stop pod and terminate. I  don't need it. So I am deleting everything. Okay,  

• 00:17:42 one more time I am going to test. So the  Jupyter Lab and the Jupyter Lab. I hope  

• 00:17:48 one of them starts. Yes, this one started. So  you see this is this one and this one didn't  

• 00:17:54 start. So I am going to delete that one, which is  this one. So stop the pod and terminate the pod.

• 00:18:01 Then I am going to upload files. First of  all, I am going to install the ComfyUI. So  

• 00:18:07 upload the ComfyUI zip file into here and let's  refresh. Okay, extract archive. By the way, you  

• 00:18:16 need to wait for upload to be completed. It is not  completed yet. So the extract would not work. Yes,  

• 00:18:22 it didn't work. Okay, it is completed. Now right  click and extract archive. Refresh. Yes. Then open  

• 00:18:30 RunPod instructions .txt file. Select this install  command, terminal, copy paste and that's it. You  

• 00:18:37 are ready. You just need to wait for installation  to be completed right now for ComfyUI.

• 00:18:44 The installation speed on RunPod 100 percentage  depends on the pod you got. Sometimes it can take  

• 00:18:50 one hour, sometimes it can take five minutes.  Extremely undependable and unpredictable. On  

• 00:18:58 Massed Compute, it is always fast, but on RunPod,  it is up to your chances. Moreover, if you use the  

• 00:19:04 permanent storage, it is way slower than getting a  new pod as I did. But permanent storage advantage  

• 00:19:11 is that you can start multiple machines on the  same storage and it is always kept as it is.  

• 00:19:18 This machine looks like a decent speed. It started  installation sooner than I expect, but let's see.

• 00:19:24 Okay, so the ComfyUI installation has been  completed. Now we will install SwarmUI.  

• 00:19:30 Upload the SwarmUI zip file into your workspace.  Wait for upload to be completed. It is uploading  

• 00:19:36 right now. And what if if you want to just use the  ComfyUI? You can use it. We have the instructions  

• 00:19:43 all here. So the upload has been completed. You  see SwarmUI model downloader, extract archive.  

• 00:19:50 Then it will extract like this. Find the RunPod  SwarmUI install instructions. Then copy this  

• 00:19:57 entire string like this and open a new terminal.  If you get this, just ignore. Ignore. This is due  

• 00:20:04 to internet connection. Okay, now it's fixed. And  hit enter. Now we will install SwarmUI instantly.  

• 00:20:12 You will see that because we have installed the  ComfyUI. Okay, just wait. It will give us a secure  

• 00:20:20 Cloudflare link to connect SwarmUI. You can also  use the RunPod proxy, but I don't recommend it  

• 00:20:27 because it is very problematic and slow. Okay, we  are still waiting. Yes. Now the SwarmUI started  

• 00:20:34 and we got the Cloudflare URL here. Let me zoom in  to show you. So this is the Cloudflare URL. Click  

• 00:20:43 it and the SwarmUI interface will start like this.  Agree, customize, important. You can choose any,  

• 00:20:50 next. This one, next. This is none. We don't  install ComfyUI local. Important, don't forget,  

• 00:20:57 none, next. I don't want to download anything.  Next, and yes, I am sure, and ready. You see?

• 00:21:04 Now we are going to add two backends because we  have two GPUs. Click the ComfyUI self-starting.  

• 00:21:10 Okay. It shows double, but it is actually one.  Let's just refresh. Go to server backends,  

• 00:21:15 yes. And add another one. And that's  it. Now our ComfyUI is installed here.  

• 00:21:22 Enter inside it. You will see that there is  main.py. Right click and copy path. These  

• 00:21:28 errors are not important, they will get  fixed. Click this icon, copy paste it here,  

• 00:21:33 but put a backslash to the beginning because  all RunPod paths start with this backslash.  

• 00:21:40 I'm going to use Sage Attention. If you  get black output, remove this, but you  

• 00:21:45 shouldn't. Then make this OverQueue 0 and save.  We will do the same in the here. So copy paste,  

• 00:21:53 copy paste. Make this GPU ID 1 and make this 0  and save. Now wait for backends to start. It is  

• 00:22:00 starting. It should be fairly fast. Meanwhile  waiting this to make it ready, you see we have  

• 00:22:07 no models yet. So we are going to use the SwarmUI  downloader inside workspace. Find the RunPod model  

• 00:22:14 download instructions .txt file, open it, copy  this part, copy, control C, open a new terminal,  

• 00:22:22 paste it and hit enter. This will start the  model downloader where you can download our  

• 00:22:27 bundles, any single model, anything. I have  explained everything in this tutorial video,  

• 00:22:33 so watch it. I am not going to show everything  again. That tutorial is mandatory. And click  

• 00:22:38 this link and you will get the model downloader  interface. It will automatically recognize our  

• 00:22:44 model folder. Let's go to SwarmUI bundles and  download the image generation models. But you  

• 00:22:51 can download any bundle or any single model.  Everything is possible. You can just search,  

• 00:22:57 it will list you all the models. You can use  the URL downloader, everything is fine. Let's  

• 00:23:01 see the download speed. Yes, really good, like  600 megabytes per second. This is amazing speed  

• 00:23:08 and our SwarmUI is still loading the backends. You  can go to logs, debug and see what is happening,  

• 00:23:14 but it should become available very soon. If this  doesn't start for any reason, just stop your pod,  

• 00:23:22 start again or restart and run the commands again,  it should work. But I think it will work right  

• 00:23:27 away. It is just downloading few models. And our  models are getting downloaded. My downloader is  

• 00:23:33 doing hash check. It is extremely robust. So when  you watch this tutorial, you will see everything.

• 00:23:38 Okay, so the model downloads have been completed.  Let's go back to our SwarmUI. Let's click refresh.  

• 00:23:45 The models should appear. Yes, models appeared,  as you can see. Then let's go to presets.  

• 00:23:51 Currently we don't have any presets, so  we need to import them. Import presets,  

• 00:23:56 choose file. In your extracted folder, right click  the Swarm model downloader and let's extract it.  

• 00:24:04 I'm using the WinRAR, you can use anything.  Extract files. Okay. Yes to all. Enter inside  

• 00:24:10 the extracted folder and select the amazing Swarm  preset latest version. You can click overwrite and  

• 00:24:16 import and it will import all of our presets.  Then click refresh and sort by name and ready.  

• 00:24:24 So the rest is exactly as in the Windows tutorial  part, but let's make a demo. So I will select the  

• 00:24:31 Wan 2.2 generate realistic images. So reset  params to default and direct apply. Photo of  

• 00:24:37 a an old wise man. Let's generate 10 images  so we can see the speed of both GPUs. First  

• 00:24:44 it will load models to the GPU. This takes time  on RunPod, so we need to wait. You can watch the  

• 00:24:52 logs and what is happening. Moreover, it is  still waiting for connecting to the server.  

• 00:24:59 RunPod is always very slow compared to Massed  Compute. Okay, we are still waiting. Yeah,  

• 00:25:05 we are waiting. Let's click generate again. It  should see our command on the logs, but I still  

• 00:25:12 don't see. Let's copy this. Okay, it started. You  see? 20 current generations because we clicked  

• 00:25:19 double times the generate, but it is slow.  You just need to wait. Now in the debug, yes,  

• 00:25:25 now it's starting to load models. Always patiently  wait, it gets faster once the models loaded.

• 00:25:33 Okay, so the generations started. We can see the  previews. For example, this is the image. The rest  

• 00:25:40 is exactly as in the Windows tutorial part. You  can of course always apply the image upscale which  

• 00:25:47 improves the quality significantly. To download  these images, you can click more and download or  

• 00:25:52 go back to your workspace inside SwarmUI, inside  output. You can download from here. Right click  

• 00:25:59 and download as an archive. It will download all  the generated images and videos. If you want to  

• 00:26:05 use these presets, you just need to download their  models. In our model downloader, we have bundles.  

• 00:26:10 You see Swarm bundles, image generation bundle,  Qwen image core bundle, Wan 2.2 core bundle. For  

• 00:26:16 generating videos, I recommend this. Wan 2.1 core  bundle, flux models core bundle. Once you download  

• 00:26:22 those bundles, these presets will become available  to use to you. And that's it. I hope you have  

• 00:26:28 enjoyed. Don't forget to turn off your machine  once you are done. It's inside pods. So here the  

• 00:26:37 machine, I can turn it off from stop pod and I  can start again. This is one of the advantage of  

• 00:26:43 RunPod. All the data will remain. After starting  again, I just need to start the SwarmUI same as I  

• 00:26:50 have installed. It will be instant and all will  be ready. However, you see this machine will  

• 00:26:55 use 6 cents per hour until I terminate it. When I  click the terminate pod, it will be gone forever.  

• 00:27:04 For example, this pod is running on my storage  from here. So in this pod, there is no stop pod.  

• 00:27:10 I can terminate it, but all the data will remain  in my storage. It's exactly same logic. You just  

• 00:27:16 make a new network volume and the rest is same.  So hopefully see you later. Thank you so much.