SwarmUI free Kaggle Account Notebook Full Tutorial - SD 1.5, SDXL, SD 3.5, FLUX, Hunyuan, SkyReels

[FurkanGozukara]

SwarmUI free Kaggle Account Notebook Full Tutorial - SD 1.5, SDXL, SD 3.5, FLUX, Hunyuan, SkyReels

Full tutorial: https://www.youtube.com/watch?v=VR1s7LxK5ZU

You want to use latest generative AI models but your computer is weak and you don't want to pay money to online services? Then you are at the right tutorial. By utilizing the most advanced UI, the SwarmUI, we are going to use the very best and latest generative AI models on a free Kaggle account. Kaggle provides 2x T4 GPU for free for 30 hours every week. With SwarmUI and my specially crafted notebook and free GPUs of Kaggle, you can use Stable Diffusion 1.5 (SD 1.5), Stable Diffusion XL (SDXL), Stable Diffusion 3 (SD 3), Stable Diffusion 3.5 - Medium and Large, FLUX DEV, FLUX Fill (inpainting), FLUX Redux, FLUX Canny, FLUX Depth, Hunyuan text-to-video, SkyReels text-to-video and image-to-video, Mochi text-to-video and many other generative AI models easily.

🔗 Kaggle Notebook Shared Post (the one used in the tutorial) ⤵️

▶️ https://www.patreon.com/posts/click-to-open-post-used-in-tutorial-106650931

🔗 SwarmUI Mega Tutorials Playlist ⤵️

▶️ https://www.youtube.com/playlist?list=PL_pbwdIyffsku5Rfv2CJZSiFdGXtkkdiB

🔗SECourses 10000+ Members Discord ⤵️

▶️ https://discord.com/servers/software-engineering-courses-secourses-772774097734074388

🔗SECourses Amazing Generative AI GitHub ⤵️

▶️ https://github.com/FurkanGozukara/Stable-Diffusion

🔗SECourses AI APPs Index ⤵️

▶️ https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/Patreon-Posts-Index.md

🔗RTX 5090 Benchmarking Video Series Playlist ⤵️

▶️ https://www.youtube.com/playlist?list=PL_pbwdIyffslNd9aLizjQtHVHAMA6tpfT

00:00:00 Introduction to the tutorial

00:00:16 How to download Kaggle notebook and load it into Kaggle and start installation

00:01:06 Kaggle configuration before starting the notebook

00:01:24 How to start Kaggle session and start installation of SwarmUI on a free Kaggle account

00:02:50 How to understand Kaggle cell execution / installation has been completed

00:03:11 How to clear cell execution outputs / logs

00:03:42 How to disable of downloading certain models in given cell / preventing that line execution

00:03:53 How to check how much disk space you have on Kaggle and how much left to use

00:04:11 Some of the models we preset put into Kaggle

00:05:10 How to learn more information about SwarmUI and how to use it

00:05:42 How much model download speed we are reaching with specially crafted downloader for Kaggle

00:06:01 How to install and start SwarmUI after download of models have been completed

00:07:03 When do we click visit site to access SwarmUI

00:07:28 Which options you need to select when installing SwarmUI on Kaggle

00:08:19 How to set backends of SwarmUI and change root folder path on Kaggle

00:09:08 How to see detailed logs of what is happening in debug menu of SwarmUI

00:09:18 How to set root folder path to Kaggle temp path we are using - mandatory

00:09:53 How to start using after SwarmUI is ready and set parameters of Hunyuan fast video model

00:12:26 How to stop SwarmUI, load a new model, and start SwarmUI again

00:14:02 Restarting SwarmUI on Kaggle

00:15:28 Generating an image with a SDXL model as an example

00:16:40 Generating a video with fast Hunyuan Q4 model

00:16:57 How to set Sigma Shift parameter

00:19:45 How to download SwarmUI generated videos and images on Kaggle as a single zip file

Below songs used with -40 sound level

0 Song: Cartoon, Jéja - Why We Lose (feat. Coleman Trapp) [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/whywelose

Watch: http://youtu.be/zyXmsVwZqX4

1 Song: Unknown Brain - Superhero (feat. Chris Linton) [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/superhero

Watch: http://youtu.be/LHvYrn3FAgI

2 Song: Unknown Brain - MATAFAKA (feat. Marvin Divine) [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/MATAFAKA

Watch: http://youtu.be/AlXfbVpDUdo

3 Song: Robin Hustin x TobiMorrow - Light It Up (feat. Jex) [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/LightItUp

Watch: http://youtu.be/bdE_SyHad90

4 Song: Egzod, Maestro Chives, Neoni - Royalty [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/Royalty

Watch: http://ncs.lnk.to/RoyaltyAT/youtube

5 Song: Lost Sky - Where We Started (feat. Jex) [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/WhereWeStarted

Watch: http://youtu.be/U9pGr6KMdyg

6 Song: Dirty Palm - Freakshow (feat. LexBlaze) [NCS Release]

Music provided by NoCopyrightSounds

Free Download/Stream: http://ncs.io/Freakshow

Watch: http://youtu.be/2jwj9wVx3mg

Video Transcription

• 00:00:00 Greetings everyone. In today's video, I am going  to show you how to use SwarmUI on a free Kaggle

• 00:00:07 account, so that you don't need to pay to any  online services. This is especially useful if

• 00:00:14 you don't have a powerful GPU. The link of this  post will be in the description of the video. Just

• 00:00:20 download the latest installer, you can download it  from here. Then go to Kaggle, register an account

• 00:00:26 and verify your phone number. This is mandatory.  After you have verified your phone number, click

• 00:00:32 create, new notebook. Wait for editor to be loaded  like this. Once you get this screen, click file,

• 00:00:41 import notebook. Then you can drag and drop the  downloaded file to here, or you can click browse

• 00:00:47 files and select the downloaded notebook like  this. Then click import. Once you see the file

• 00:00:56 has been imported like this, you can click X icon  here. You don't need to wait fully, but you can

• 00:01:04 also wait, it's an option. After that, verify that  accelerator is selected as GPU T4 x2 and internet

• 00:01:15 is on. This is mandatory, otherwise it will not  work. Then we will start our session. This is

• 00:01:21 the first cell that we need to install, so let's  start our session from here. Wait for session to

• 00:01:29 be started. Once you see a green circle here, you  see, it says running. Click here and verify that

• 00:01:37 you have the GPU and everything is working as  expected like this. Then click this cell. When

• 00:01:46 you click this cell, you can either click here  to play it, or you can click here to play it. You

• 00:01:51 see run current cell. But make sure that you have  selected the accurate cell. Then let's click this,

• 00:01:58 and it will start installation of the necessary  libraries to run the SwarmUI on a free Kaggle

• 00:02:04 account. With this SwarmUI, you can use FLUX Dev  model, SDXL model, Stable Diffusion 1.5 model,

• 00:02:12 Stable Diffusion 3.5 model, Hunyuan video model,  Mochi video model, whatever SwarmUI supports, you

• 00:02:22 can use it here, as long as it doesn't exceed the  requirements. Because we have only 29 gigabytes

• 00:02:32 of RAM and we have GPUs that are 15 gigabytes.  However, since we have two GPUs here, we can

• 00:02:40 generate two images or two videos at the same  time, as long as they are fitting into GPU and

• 00:02:47 they are not causing out of RAM error. So you see,  this cell execution has been completed because

• 00:02:56 there is not anymore running icon here. You can  also see the status from here as well. There is

• 00:03:02 no running icon anymore. Then scroll down. You can  see the installation logs like this when you are

• 00:03:09 scrolling down. Or you can right click and you  can say clear outputs of all cells, so that it

• 00:03:18 will clear all the installation logs. After that,  go to the Step 1 cell. So we have a lot of cells

• 00:03:27 to automatically download the models. The favorite  models that you might enjoy. Like this cell will

• 00:03:35 download SDXL models or Stable Diffusion  1.5 models. We have set some models for you

• 00:03:41 to automatically download. If you don't want to  download a certain model, you can put a hashtag in

• 00:03:47 front of that model and it will disable download  of that model. Since we have limited space on

• 00:03:56 Kaggle, you see we have 57 gigabytes of disk space  total, we cannot download every model. You need to

• 00:04:03 pay attention to the models that you need, that  you are going to use after the installation. So

• 00:04:08 download the models accordingly. For example, this  cell will download the very best deterministic

• 00:04:15 upscale models. This cell will download a better  clip model for the FLUX. This cell will download

• 00:04:21 FLUX Dev model, Stable Diffusion 3.5 large model.  This cell will download the FLUX Redux model. So

• 00:04:30 read every one of the cells and decide which ones  you want to download according to your needs. For

• 00:04:38 example, this cell will download the Hunyuan text  to video model. This is higher quality model, 20

• 00:04:46 steps. It is going to download Q6 GGUF model. This  cell will download faster Hunyuan video model. This

• 00:04:57 is 6 steps model. So read all of the cells and  download whichever model you need. I am going to

• 00:05:05 download this fast model to generate some example  videos. We already have amazing SwarmUI tutorials

• 00:05:13 playlist, so you can look at our playlist and  watch all of our SwarmUI tutorials to learn more

• 00:05:22 about SwarmUI. Hopefully, I will put the link into  the description of the video, so you can quickly

• 00:05:28 navigate to there to learn how to use SwarmUI with  different models and its different features. Once

• 00:05:35 you have installed, everything is exactly same to  use on Kaggle. So the download was extremely fast.

• 00:05:42 You see 678 megabytes per second. This is also an  optimization that I do. So it took 27 seconds to

• 00:05:52 download. So after you have checked everything and  you are ready to use SwarmUI, you downloaded all

• 00:05:59 the models. We are going to use this to be able  to connect the SwarmUI interface. So click this

• 00:06:06 link. Register a free account. You can register  any account. Then click to copy your token like

• 00:06:16 this. Then you will see put your token here, put  your token there. Then when this cell is selected,

• 00:06:24 click play icon once. You can run from here or  from here. Let's run from here. Then you will get

• 00:06:31 a link like this. Open this link by clicking  that and you will get to this page. However,

• 00:06:37 do not click visit site yet. We only click visit  site when the application is running at localhost.

• 00:06:46 I will show you what I mean. Get to Step 3 cell. This cell will install the SwarmUI and

• 00:06:53 start it. So click to run that cell and just  wait until you see this. Let me show you what you

• 00:07:02 are going to see. You are going to see starting  web server on http://0.0.0.0:7801. This means that the

• 00:07:13 application started locally on the Kaggle server.  However, to be able to access it, we need a public

• 00:07:20 endpoint, which is this link. Now click visit site  and the SwarmUI interface will load. On Kaggle,

• 00:07:29 we are using pre-installed ComfyUI backend,  which we have installed at the very first cell

• 00:07:37 that we have run. So click agree. Click customize  settings. This is super important. Click next.

• 00:07:44 Select the theme you want. Next. Just yourself on  this PC. Next. And this is crucial, none. Select

• 00:07:52 none because we are going to use pre-installed  ComfyUI backend. So this is important. Select

• 00:07:59 none and click next. Do not download any model  because we have downloaded everything we needed

• 00:08:05 into the temp disk because it is the bigger disk.  Then click next. Then yes, I am sure install now.

• 00:08:13 It will be almost instant like this because we  didn't install any backend and it is done. Now,

• 00:08:20 we are going to add backends. So click server,  click backends and click ComfyUI self starting.

• 00:08:29 Click okay. Then click ComfyUI self starting  again and click okay. Then go back to the Kaggle

• 00:08:36 notebook editor and go back to the top of the  Step 3 cell. You will see that there is this

• 00:08:44 text after SwarmUI started first add. So select  this part and copy it. Go back to the SwarmUI

• 00:08:51 interface. Paste it here. Then make the GPU ID  of the second one as GPU ID 1. Then click save

• 00:09:01 and click save. Now it will start the backends,  install the missing libraries or anything. You can

• 00:09:09 go to logs, click debug and watch the progress.  There is one another thing that we have to make

• 00:09:16 which is server configuration. Click server  configuration, return back to Kaggle and you see

• 00:09:22 there is this folder, Kaggle temp. Copy it. Return  back to SwarmUI interface. Paste it here. You

• 00:09:29 see model root. Once you paste it and click it  another cell, you will see that there is save.

• 00:09:35 So click save. So now it will change the model  root and it will load the backends. We don't need

• 00:09:42 to restart SwarmUI multiple times anymore because  this will work perfectly now. Just go to logs and

• 00:09:48 watch in the debug and you see loading backends  disappeared. So the SwarmUI is ready to use. So

• 00:09:56 click models and click refresh. Currently we only  downloaded fast Hunyuan video because everything is

• 00:10:02 same as in other tutorials, just watch them.  But I will make an example of Hunyuan video. So

• 00:10:08 click this. And once you clicked it, you see it is  selected. This is a fast model so it is 6 steps.

• 00:10:16 But before doing that, we need to install a node,  if I remember correctly. So click this hamburger

• 00:10:23 menu and click load now. And you see it is asking  me to install GGUF support. So install and okay.

• 00:10:33 Just wait for installation to be completed. You  see it says backends are still loading because

• 00:10:38 it installed the necessary library and it is  reloading the backend and now it is done. Then we

• 00:10:45 can continue setting up the necessary parameters.  This is CFG 1 and click display advanced options. I

• 00:10:52 am going to make another tutorial for Hunyuan video,  but I am showing here quickly so that you will see

• 00:10:58 there is text to video now. You can set number of  frames, FPS, another thing. Let's just generate

• 00:11:04 25 frames, so it will be one second video. And  it is 6 steps, CFG scale 1. In the sampling,

• 00:11:13 we are going to make the FLUX guidance scale  10. This is for fast Hunyuan video and we are

• 00:11:19 also going to change something else which  is in the advanced sampling and Sigma Shift,

• 00:11:26 it is 17. I will use an example prompt that I  have found on the internet. This is an example

• 00:11:33 prompt for Hunyuan video. You can read this, just  pause the video and read it and click generate.

• 00:11:41 Now let's watch how much time it is going to take.  I hope that it won't take too long because we have

• 00:11:47 limited VRAM. So first of all, it will download  the necessary missing models, then it will start

• 00:11:54 generation. You can always watch the status from  the debug menu. So it is downloading the LLAVA LLAMA3

• 00:12:01 FP8 scaled version automatically. You can  also monitor some of the progress on the Kaggle

• 00:12:09 notebook as well. So let's just wait. Okay, so the  model has been downloaded and the video generation

• 00:12:16 started. However, there is an error, the mistake,  which is that it is using our entire GPU memory,

• 00:12:26 therefore it is just too slow. So what I am  going to do is, I have replaced the download

• 00:12:32 with Q4 instead of the Q6, so it will use lesser  VRAM, therefore it should work faster. So how

• 00:12:40 am I going to do that? By the way, in your case,  when you download the notebook, you will get the

• 00:12:46 updated Kaggle notebook, not the Q6 version.  But you will learn how to cancel and run the

• 00:12:53 application again when you want to download  a new model or make changes. So cancel run,

• 00:13:00 and the run has been cancelled. Let's redownload  the Q4 version. So I will click this and it should

• 00:13:06 redownload very fast. Yes, we can see the speed  here. So it is around 700 megabytes. Let me also

• 00:13:15 update the notebook. Okay. And you can watch  the disk space that you are using, so you won't

• 00:13:23 get out of disk space. Okay. Disk space may get  updated a little bit later. Let's also download

• 00:13:31 some of the SDXL models, so I can also show you  how to use them. The SDXL model download was step

• 00:13:37 1, so I will run this cell. It will download the  SDXL models extremely fast. Just wait until cancel

• 00:13:46 run disappears. Okay, new downloads have been  completed. Now we are using around 50 gigabytes

• 00:13:54 of disk space. Now we will restart the SwarmUI.  But we don't need to reinstall. How to restart?

• 00:14:02 Go to Step 2 cell again, run it. Get a new link  here, open it. But do not click visit site yet.

• 00:14:10 Scroll down to the Step 4 and just execute it.  And when you see that starting web server on this

• 00:14:18 local URL, you are ready. Click this cell. Go  to the link and visit site. Okay. It is asking

• 00:14:26 us to reinstall, this shouldn't happen normally.  However, on Kaggle, this can happen sometimes,

• 00:14:32 so if you encounter that, don't get confused. So  we are going to reinstall quickly as before. Next,

• 00:14:42 next. Okay. So it should be instant. Yes. The  bad side of this is that we need to re-add the

• 00:14:51 backend. So we will do that. Unfortunately, on  Kaggle, it is not as easy as using in our own

• 00:14:59 PC. Scroll back top of the Step 3. Copy the  backend link, this one, pay attention. Copy paste

• 00:15:08 here and here. Set the GPU ID, save, save. Go back  to server configuration. Copy the folder path like

• 00:15:17 this. Change here and save and wait for backends  to be loaded. You can monitor logs, debug. Okay,

• 00:15:26 it is loaded. Go to back to the generate and click  models, refresh. Don't forget to refresh. Now we

• 00:15:34 have different models. Let's generate an image  with the Realistic Vision XL version 5. This is

• 00:15:40 an SDXL model. Let's say super fast car. Then  I will try Hunyuan Q4 GGUF model. First, it will

• 00:15:51 load the model, then it will start generation.  You can watch logs in here in the server logs.

• 00:15:59 This should be pretty fast compared to the Hunyuan  video. You can watch the status here also. You see

• 00:16:06 1 current generation, 1 queued, 1 waiting  on model load. Okay, model loaded and generation

• 00:16:13 has been started and we should get the results  in a few seconds. We can see the speed here.

• 00:16:19 It is 1.23 IT per second and the image has been  generated. Yes. You see it is working perfect.

• 00:16:28 Now let's try again the fast Hunyuan. By the way, it  still shows the Q6 because we had downloaded it,

• 00:16:37 but we are going to select Q4. Okay, let's select  it. We need to reset the parameters, CFG 1, 6

• 00:16:46 steps. Let's click load model. Okay. You see this  time it didn't ask me to reinstall the necessary

• 00:16:54 node because it was installed. Then click display  advanced options. We are going to select the

• 00:17:00 sampling and FLUX guidance scale as 10 and we  are going to set the Advanced Sampling Sigma

• 00:17:07 Shift 17. And let's copy paste our prompt. Okay  and generate a video. Let's see how it will work.

• 00:17:18 You can also change the video format from here to  GIF, WebM or MP4. This will change the extension,

• 00:17:28 the video file format that it is going to  generate. Let's go to server and let's see. Okay,

• 00:17:35 it gave some errors, but I think it is fixing  itself. Let's see. I think it is getting loaded

• 00:17:42 right now. Yes, you can also see the status here.  You see the RAM is increasing. That means that it

• 00:17:49 is loading the model into the RAM first, then it  will load it into the GPU memory. Okay, this time

• 00:17:56 it loaded it into our second backend into the GPU  ID 1, you can see that. It is only using 11.9

• 00:18:04 gigabytes of VRAM memory right now, so the speed  should be relatively faster than the Q6 model.

• 00:18:13 However, still, the GPUs on the Kaggle is slow,  therefore it will take time. And as you generate

• 00:18:19 a bigger frames having video, it will take even  longer, or if you generate a bigger resolution,

• 00:18:26 it will take longer. This is just a demonstration  how it works on Kaggle. How to use the SwarmUI,

• 00:18:34 watch the tutorials and I am hopefully going to  make another tutorial, a more detailed tutorial

• 00:18:39 for Hunyuan video. Okay, so the video generation has  been started. However, it is extremely slow due

• 00:18:47 to the GPU of the Kaggle. One step is taking 900  seconds. So if you are not in hurry, you can of

• 00:18:56 course generate video and wait it to be completed  because Kaggle provides 30 hours of free GPU time

• 00:19:05 every week. We can see that, let me show you.  Okay, you see that I have left 28 hours of GPU

• 00:19:15 time for this week. This is weekly and for free.  However, it is just extremely slow for this model,

• 00:19:22 but you can use it for SDXL, FLUX Dev, Stable  Diffusion 3.5, SD 1.5, even Mochi video model.

• 00:19:33 So currently this is the speed, it may get faster  when you are using this tutorial, but you know,

• 00:19:39 I am teaching you how to use and the rest is up  to you to figure it out how you want to use it.

• 00:19:45 So how you can download the generated videos  or images? Once you are done with everything,

• 00:19:53 click cancel run here. Then go to the very bottom  cell and click this play icon. This is usually the

• 00:20:01 way I develop my Kaggle notebooks and you see it  has generated a zip file. This zip file will be

• 00:20:08 here. Click this icon, it will list then you see  there is generated zip. So click here and download

• 00:20:16 and it will download all the generations you have  made on Kaggle as a single zip file. It will make

• 00:20:23 your life easier. Hopefully see you in the future  videos. Please subscribe. I will keep updating and

• 00:20:31 improving this notebook for SwarmUI and if you  have any questions, just reply to the Patreon

• 00:20:38 post or this video and hopefully see you later.  Don't forget to join our Discord channel as well.