Skip to content

How To Use Storage Network Volume In RunPod and Latest Version Of Automatic1111 With All ControlNet

Jump to bottom
FurkanGozukara edited this page Oct 21, 2025 · 1 revision

How To Use Storage Network Volume In RunPod & Latest Version Of Automatic1111 With All ControlNet

Full tutorial link > https://www.youtube.com/watch?v=8Qf4x3-DFf4

How To Use Storage Network Volume In RunPod & Latest Version Of Automatic1111 With All ControlNet

image Hits Patreon BuyMeACoffee Furkan Gözükara Medium Codio Furkan Gözükara Medium

YouTube Channel Furkan Gözükara LinkedIn Udemy Twitter Follow Furkan Gözükara

In this tutorial I have shown how to login your RunPod account and how to deploy a Pod with any specific template in the newest updated RunPod interface for both secure and community cloud. Moreover, how to use Storage Network Volume system of RunPod to get permanent storage that can be mounted to any available GPU. Furthermore, how to terminate permanent storage and start again on a new pod. In addition, how to deploy multiple GPUs on a single pod. Additionally, how to 1 click download all of the ControlNet models, install necessary libraries for InstantID and IP Adapter Face ID, and download best SDXL and SD 1.5 models. Moreover, how to terminate auto started Web UI and verify 0% VRAM is used before starting any training. Furthermore, I have shown how to use InstantID face transfer to obtain amazing stylized images instantly.

RunPod / Linux : 1 Click Installer for Automatic1111 SD Web UI, SDXL, ControlNet, All ControlNet Models, TensorRT (RTX Accelerator) For RunPod / Any Linux System ⤵️

https://www.patreon.com/posts/86438018

Windows : 1 Click Installer for Automatic1111 SD Web UI, SDXL, ControlNet, All ControlNet Models, TensorRT (RTX Accelerator) on Windows ⤵️

https://www.patreon.com/posts/86307255

Biggest & The Best How To Use RunPod Tutorial ⤵️

https://youtu.be/QN1vdGhjcRc

Biggest & The Best How To Use ControlNet Tutorial ⤵️

https://youtu.be/3E5fhFQUVLo

00:00:00 Introduction to how to use RunPod with permanent Storage Network Volume and latest version of Automatic1111 SD Web UI and ControlNet

00:01:40 How to register and login your RunPod account

00:01:54 How to deploy a pod in the newest RunPod interface

00:02:22 How to deploy a Pod on community cloud on the newest RunPod interface

00:02:39 How to select / set your RunPod template

00:02:53 How to customize deployment of RunPod template and increase storage disk size and expose proxy ports to connect

00:03:40 How to use Storage Network Volume system of RunPod to get permanent storage that can be mounted to any available GPU

00:05:10 How to setup a permanent Storage Network Volume in a specific region on RunPod

00:06:31 Why you shouldn't change template name

00:07:06 Terminate option of permanent storage vs stop pod of community cloud Pods

00:07:57 How to install latest version of Automatic1111 SD Web UI on RunPod with ControlNet models

00:08:15 How to rent multiple GPUs on the same machine

00:08:42 How to connect Jupyter Lab interface to start installing and using AI apps such as Automatic1111 SD Web UI

00:09:10 How to upload Automatic1111 SD Web UI latest version installation scripts into the Pod and install

00:09:37 How to prevent automatic restart of Web UI in the Stable Diffusion template with modifying relauncher.py file

00:10:06 How to restart your Pod after replacing / overwriting your relauncher.py file

00:10:49 How to execute installation commands

00:11:26 Which models will install all sh file download

00:12:25 Why getting permission denied error is not important and how we are terminating automatically started Automatic1111 SD Web UI

00:12:36 Why it is important to verify you terminated auto started Automatic1111 SD Web UI and thus have 0% VRAM usage before starting any training

00:13:02 What kind of download speeds we are getting in RunPod community cloud and secure cloud pods

00:14:09 How to use bidding / spot bidding option of RunPod to get even much more cheaper GPUs / Pods

00:14:39 If someone else overbids you what happens when you are using spot bidding

00:15:57 How to close your storage network volume or regular pod and start using again

00:16:16 What if a GPU is not available

00:16:26 How to completely delete a pod - all data is gone

00:16:52 How to deploy storage network volume into a better GPU

00:17:30 How to run Automatic1111 SD Web UI on RunPod after stopping and starting Pod again or restarting the Pod

00:17:45 Why moving storage network volume to a new Pod takes time

00:18:36 Why manually starting your Web UI after terminating older automatically started one is more advantageous - you can watch the terminal

00:18:49 Why I have removed the skip install argument from the Automatic1111 Web UI arguments

00:18:59 How to install other and newer extensions on RunPod Automatic1111 Web UI

00:19:19 How to connect back to the started Automatic1111 SD Web UI interface with http proxy port

00:19:31 How to enable Gradio live share on RunPod to use Automatic1111 SD Web UI on any device such as your phone, tablet or on another computer

00:19:42 How to edit starting arguments of Automatic1111 on RunPod

00:20:50 We got connected to the Web UI interface and checking all available ControlNet models

00:21:04 Example usage case of ControlNet

00:22:17 How to use recently added InstantID in ControlNet in Automatic1111 SD Web UI

00:24:32 The quality difference and easiness of using of our own developed InstantID Gradio app

00:26:26 Testing image on my own developed advanced Gradio app for InstantID

00:26:52 The list of amazing features of our Gradio app for InstantID

00:27:48 What happens if your GPU starts using shared VRAM

00:29:00 How to get best results with InstantID

Video Transcription

  • 00:00:00 Greetings, everyone.

  • 00:00:01 In this tutorial, I will show you how to install and update the latest Automatic1111 Web UI

  • 00:00:07 on RunPod.

  • 00:00:08 This includes also Linux systems if you have a local Linux because RunPod uses Linux.

  • 00:00:14 You see version 1.7.0 with the latest libraries.

  • 00:00:19 Moreover, I will show how to install automatically TensorRT.

  • 00:00:23 I will show you how to also download and install all of the ControlNet models.

  • 00:00:29 Let me show you them.

  • 00:00:30 So when I select none here, let's say all and let's refresh.

  • 00:00:35 All of the models of ControlNet are automatically downloaded.

  • 00:00:39 Everything will be one click for you.

  • 00:00:41 So with following this tutorial, you will be able to use the latest Automatic1111 Web

  • 00:00:45 UI on RunPod or any Linux system, any cloud service that supports Linux with the convenience

  • 00:00:52 of downloading all of the ControlNets automatically, including even the InstantID and IP adapter.

  • 00:01:00 It will also install the necessary libraries for them to work or the InstantID.

  • 00:01:05 You will be able to use InstantID as well.

  • 00:01:09 Let me show you the model.

  • 00:01:10 Yes, here, for example, this one or this one.

  • 00:01:13 So this tutorial is the perfect tutorial that you need if you want to use cheap GPUs on

  • 00:01:20 RunPod with Automatic1111 Web UI.

  • 00:01:23 One another thing that I'm going to show you is how to have a permanent storage system

  • 00:01:28 like this with storage network volume, how to mount it, how to reuse it.

  • 00:01:32 This will be a super tutorial.

  • 00:01:35 So let's keep watching.

  • 00:01:36 So everything that you need is shared in this post.

  • 00:01:40 We will begin with logging in into our RunPod account.

  • 00:01:45 You can follow this link.

  • 00:01:46 This link will get me some credits and I can use those credits to do research and do more

  • 00:01:53 tutorials for you.

  • 00:01:54 So this is the newest interface that we have. To deploy your pod

  • 00:01:59 click this pod icon and then click GPU pod.

  • 00:02:03 In the newest interface, you will see that automatically secure cloud is selected.

  • 00:02:08 Secure cloud is better than the community cloud sometimes, not every time.

  • 00:02:13 However, it is also more expensive and the benefit of secure cloud comes from the storage.

  • 00:02:19 I will show that too.

  • 00:02:21 So let's open first our community cloud, which is cheaper.

  • 00:02:26 Select extreme speed from here.

  • 00:02:27 Then let's pick RTX 3090, which is one of the most affordable price-performance GPUs.

  • 00:02:35 You can pick anything you wish.

  • 00:02:37 I will go with this one.

  • 00:02:39 Okay, at this interface, we are going to select our template.

  • 00:02:42 This is super important. Type here Stable Diffusion because we want to use Automatic1111 Web UI.

  • 00:02:49 Therefore, I will use RunPod Stable Diffusion template.

  • 00:02:53 Then click customize deployment.

  • 00:02:55 We are going to increase volume disk.

  • 00:02:57 You can set it anything you wish according to your needs.

  • 00:03:00 Let's set for 150 gigabytes, set overrides.

  • 00:03:04 And we also need to give a name.

  • 00:03:07 Let's say test one is a container image name.

  • 00:03:10 Okay, everything is set up.

  • 00:03:12 You see it is how cheap.

  • 00:03:14 Continue to deploy.

  • 00:03:15 Okay, you see we got an error as this machine doesn't have so let's deploy again with another

  • 00:03:21 GPU.

  • 00:03:22 I think the same settings are deployed.

  • 00:03:25 All right, and it is getting started.

  • 00:03:27 This machine has 37 gigabytes RAM.

  • 00:03:30 And these are the upload and download speeds.

  • 00:03:33 Now this machine is not permanent, its GPUs may get used when we turn it off.

  • 00:03:40 Therefore, there is another approach which is storage.

  • 00:03:43 This storage mode will give you a deployable storage that you can deploy in any available

  • 00:03:50 pod unlike the other pod setup.

  • 00:03:53 So there are regions that you need to select and in regions you can see which GPUs are

  • 00:03:58 usually available.

  • 00:04:00 Before selecting your region, you can also go to the pods, click GPU pod, and in here

  • 00:04:07 select secure cloud and look for the data center regions that you are going to deploy.

  • 00:04:13 For example, US Orion 1.

  • 00:04:16 And in here I have RTX 4090 available RTX 6000 ADA available, and all others are not

  • 00:04:24 available as you are seeing.

  • 00:04:26 So you need to find a region that has available GPUs usually so that you can deploy anytime

  • 00:04:33 you wish.

  • 00:04:34 Currently, the available GPUs are getting scarce due to, you know, accelerated AI.

  • 00:04:41 Therefore, they need to purchase more GPUs.

  • 00:04:45 Okay, so. So look at the GPUs available and pick your region.

  • 00:04:50 For this tutorial, let's take a region as let's see, yeah, all is not available.

  • 00:04:57 Okay, A5000 is available here.

  • 00:05:01 I still prefer RTX 3090. Non available.

  • 00:05:04 I wonder if there are any other settings that we can know.

  • 00:05:08 Let's go with EU-SE-1.

  • 00:05:10 So to have a storage here, I click storage, click network volume, and I select this one,

  • 00:05:17 it says this is usually available.

  • 00:05:19 These are low available.

  • 00:05:21 Yeah, let's go with this one.

  • 00:05:23 Here you set your permanent storage.

  • 00:05:26 Let's set 150 gigabytes, you cannot reduce it, but you can expand it,

  • 00:05:30 if it be necessary. Give the name you want.

  • 00:05:33 Let's say test one create.

  • 00:05:35 Now, we will deploy it to the any available GPU.

  • 00:05:39 And you can do this anytime you wish.

  • 00:05:42 So you can also deploy it onto any available GPU.

  • 00:05:45 Okay, I will deploy it on RTX A5000, which is a decent GPU, you see this is 44 cents,

  • 00:05:52 and the community cloud was 26 cents.

  • 00:05:56 Okay, let's deploy.

  • 00:05:58 And also select your template here as this one because these will be used to continue

  • 00:06:03 deploy.

  • 00:06:04 The first deployment of the network storage may take more time.

  • 00:06:08 And you see this machine has 50 gigabytes RAM, the other one is only 26 cents.

  • 00:06:14 And it is still not.

  • 00:06:15 Yeah, looks like this didn't work.

  • 00:06:18 Maybe because we changed the name.

  • 00:06:20 So let's delete this one and deploy again.

  • 00:06:23 Let's go to the pods, GPU pod.

  • 00:06:26 And here let's select community cloud, from here RTX 3090.

  • 00:06:31 Maybe we shouldn't change this one.

  • 00:06:33 Yeah, probably this was the reason why it was failed.

  • 00:06:36 So we need to keep it as it is. Set the volume disk, set overrides.

  • 00:06:41 Okay, so they should match both.

  • 00:06:44 This was my mistake and deploy and it is getting deployed as you are seeing.

  • 00:06:50 Okay, so the private network started and you see secure cloud has much higher download

  • 00:06:56 speed and upload speed and the hard drive speed.

  • 00:06:59 But these may not be very dependable.

  • 00:07:01 Sometimes I saw that community cloud is working much faster than the secure cloud.

  • 00:07:06 So this machine has terminate option, because this is a private volume.

  • 00:07:12 So when you terminate it, it just close your pod and you can deploy it to any other machine.

  • 00:07:18 However, on community cloud, we have stop option, because when we stop our pod, it

  • 00:07:23 will just stopped.

  • 00:07:25 And then when there are available GPUs, we can start again.

  • 00:07:28 So this has two options.

  • 00:07:30 When I stop it, it will not get deleted, it will stay like this.

  • 00:07:34 And then I can terminate and this will delete my data permanently.

  • 00:07:39 However, on the secure cloud, when I click terminate, you see it is gone.

  • 00:07:44 However, my data is not gone.

  • 00:07:46 It is still inside here.

  • 00:07:48 And until I delete it, it will remain as it is.

  • 00:07:51 So let's deploy it again on any available GPU like this one again.

  • 00:07:56 So just deploy.

  • 00:07:57 Okay, now time to set up.

  • 00:08:00 So let's see, let's start this.

  • 00:08:02 By the way, you see these are the available GPUs.

  • 00:08:04 If this becomes zero, you have to move your data into another pod when you are using

  • 00:08:10 community cloud or non-storage secure cloud.

  • 00:08:14 So let's start it.

  • 00:08:15 Both are getting started. With storage

  • 00:08:18 another advantage is that you can also rent multiple GPUs if they are available.

  • 00:08:23 However, with the community cloud, you have to pick your maximum number of GPUs at the

  • 00:08:30 beginning.

  • 00:08:31 So you see, let's go back to the community cloud.

  • 00:08:33 And when I am going to first time deploy, I need to select my maximum number of GPUs

  • 00:08:38 because it is not getting increased at a later time, even if you stop your pod.

  • 00:08:42 Okay, both pods are started, let's connect to their Jupyter lab, you need to wait until

  • 00:08:48 Jupyter lab becomes available.

  • 00:08:50 Sometimes these may still show orange, but it may connect probably this is one of that

  • 00:08:55 times.

  • 00:08:56 So this was our community cloud.

  • 00:08:58 And this is our secure cloud. The setup and everything is exactly the same.

  • 00:09:03 It doesn't change. In here

  • 00:09:06 I have prepared an amazing scripts for you.

  • 00:09:10 So let's download the RunPod Auto11 version one zip file.

  • 00:09:15 And let's extract it into any folder you wish because we will just use the scripts inside

  • 00:09:20 them.

  • 00:09:21 And the scripts are here.

  • 00:09:22 As you are seeing, I have prepared RunPod instructions, you can also read this, and

  • 00:09:28 you will see everything here as well.

  • 00:09:30 So let's start the instructions.

  • 00:09:32 Okay, we got everything here.

  • 00:09:34 So okay, this still didn't start.

  • 00:09:36 Let's start it.

  • 00:09:37 First of all, this template permanently restarts the web UI.

  • 00:09:43 But we don't want that to happen.

  • 00:09:45 So therefore we will override the relauncher.py file.

  • 00:09:49 So let's upload the downloaded relauncher.py file.

  • 00:09:53 Okay, I go to the downloads, and I select relauncher.py file, and it will ask me to

  • 00:09:59 overwrite this is super important overwrite.

  • 00:10:01 After overwrite, if you double click, you will see that it becomes like this.

  • 00:10:06 After this operation, you need to restart your pod.

  • 00:10:09 Let's also do this to the other one as well.

  • 00:10:12 Okay, this is still not started.

  • 00:10:14 We can see the logs.

  • 00:10:16 Let's just wait a little bit more.

  • 00:10:18 Okay, it is also starting. Then how we are going to restart. You see in here just click

  • 00:10:23 and restart pod and it will restart the pod.

  • 00:10:26 This is just one time necessary.

  • 00:10:27 We will do this on the secure pod as well.

  • 00:10:30 So it is here, you can also alternatively drag and drop like this into this folder and

  • 00:10:36 overwrite it will do the same operation.

  • 00:10:38 Then we will also restart this one as well.

  • 00:10:41 So let's just restart it.

  • 00:10:43 Okay, this pod probably restarted.

  • 00:10:45 Now for installation, what we need to do is we need to just drag and drop files here like

  • 00:10:52 this, then open a new terminal.

  • 00:10:54 Now you can read this file very carefully.

  • 00:10:57 I have separate installers and updaters.

  • 00:11:00 For example, if you just want to update Automatic1111 web UI to the latest version, you just execute

  • 00:11:06 install latest auto 1111.sh file.

  • 00:11:10 However, if you want to install everything, then we have install everything file, this

  • 00:11:16 one just copy this, open a new terminal, you can click this plus icon terminal, copy paste

  • 00:11:22 and hit enter, and it will do everything automatically for you.

  • 00:11:26 What will this do?

  • 00:11:28 This will install latest Automatic1111 web UI.

  • 00:11:32 This will install the latest ControlNet extension.

  • 00:11:36 This will install the InsightFace library so you can use IP adapter Face ID, or you

  • 00:11:42 can use InstantID.

  • 00:11:43 This will download the models, which models?

  • 00:11:47 This will download realistic vision fp16 VAE fix, playground version two, hyper realism

  • 00:11:54 version three, and this will download all of the ControlNet models available, 52 models.

  • 00:12:00 This will also download 275 amazing focus styles.

  • 00:12:06 So this is an amazing secret that I have prepared.

  • 00:12:09 Just you execute that and you just wait.

  • 00:12:12 Let's do the same for the secure cloud as well like this.

  • 00:12:17 So the difference of secure cloud is once we set it we can deploy it to any pod we

  • 00:12:22 wish and just execute and just wait.

  • 00:12:25 You will also get this permission denied, it is not important, we are killing automatically

  • 00:12:30 started web UI.

  • 00:12:32 So you can see that GPU memory usage is dropped to zero percent.

  • 00:12:36 This is also important when you are doing training. You have to kill automatically started

  • 00:12:41 web UI.

  • 00:12:42 And for that to work, you need to change the relauncher.py file here. It is super important

  • 00:12:51 because if you don't make that change, even if you kill the web UI, it will automatically

  • 00:12:56 start again.

  • 00:12:57 So then just relax and wait for everything to be installed and downloaded.

  • 00:13:02 The download speeds are amazing, currently over 100 megabytes for the community cloud

  • 00:13:09 and over 200 megabytes for the secure cloud.

  • 00:13:12 When you look closely, it will show you how many models are automatically downloaded,

  • 00:13:18 22 over 52.

  • 00:13:20 When we look at the secure cloud, we can see that 26 over 52.

  • 00:13:26 So this is extremely convenient for you to use.

  • 00:13:29 After all models downloaded, it will start installing the newest packages and the newest

  • 00:13:35 Automatic 1111 Web UI.

  • 00:13:37 You see it is also automatically installing the InsightFace, which is a library that is

  • 00:13:42 necessary for face transfer applications.

  • 00:13:46 If you also need additional extensions, just message me from Patreon, and hopefully, I

  • 00:13:51 will also include them into the installations.

  • 00:13:54 This is secure cloud.

  • 00:13:55 It is a little bit faster.

  • 00:13:57 The community cloud is still trying to catch up with the secure cloud.

  • 00:14:02 But remember, secure cloud is 44 cents and community cloud is 26 cents per hour.

  • 00:14:09 Okay, another option, which is even cheaper, is bidding in the community cloud.

  • 00:14:15 So let's click GPU pod. In here

  • 00:14:17 let's select community cloud extreme, and let's go.

  • 00:14:20 Let's say A6000 GPU or maybe 3090 GPU.

  • 00:14:27 Let's go with 3090. In here

  • 00:14:29 you see on-demand non-interruptible.

  • 00:14:32 If you make it spot bidding interruptible, it drops to 19 cents per hour.

  • 00:14:37 This is ultra-cheap.

  • 00:14:39 In this mode, if someone overbids you, let's say they selected on-demand, and there is

  • 00:14:46 no other available GPU, so they will take away your GPU, and you will be out of GPU on your

  • 00:14:53 pod, your pod will continue working.

  • 00:14:55 However without GPU, so you will get errors

  • 00:14:58 if you are doing a GPU-dependent work. You will not lose any data unless you are processing

  • 00:15:04 some data, and you have to save at that time or something.

  • 00:15:08 So this is another cheaper option.

  • 00:15:10 Okay, everything is getting installed on both of the pods.

  • 00:15:15 You see version 1.7.0 installing the latest requirements.

  • 00:15:19 The same is happening here as well.

  • 00:15:22 You will also get this dependency message.

  • 00:15:25 Don't worry, it will work.

  • 00:15:27 Currently, it is installing the necessary libraries for TensorRT.

  • 00:15:31 TensorRT is another amazing feature and it speeds up the image generation about 70% without

  • 00:15:39 any quality loss.

  • 00:15:41 70%.

  • 00:15:42 However, for each model, you have to compile a TensorRT, but it worths if you are going

  • 00:15:48 to generate hundreds or thousands of images, it is extremely useful.

  • 00:15:52 I have a full tutorial for that the links are here and on the channel.

  • 00:15:57 Let's say you want to close your server and use it again.

  • 00:16:01 So to do that, first of all, if you are using the community cloud or private cloud without

  • 00:16:07 storage, you need to have a GPU available.

  • 00:16:11 So when I click stop, it will stop this server.

  • 00:16:15 This is the community cloud.

  • 00:16:16 Then if I want to start again, I just click here start and select the GPU if available.

  • 00:16:21 If not available, I won't be able to.

  • 00:16:23 If I want to completely delete it, I have to use terminate.

  • 00:16:26 If you use terminate on this pod, it will delete all of your data; all will be gone.

  • 00:16:32 What about the storage that we have?

  • 00:16:34 You will click this terminate, but it will not terminate the pod; it will not delete

  • 00:16:40 your data.

  • 00:16:41 It says this will remove your pod configuration.

  • 00:16:45 So let's click yes, and it is gone.

  • 00:16:47 But the all data remains here.

  • 00:16:49 Now I can deploy again into any available pod.

  • 00:16:52 I can also choose a better GPU from here.

  • 00:16:56 For example, I can deploy on an A6000 like this, and I will start using it with my all

  • 00:17:02 data.

  • 00:17:03 However, if I delete this pod, this is the community cloud, all the data will be gone

  • 00:17:09 forever.

  • 00:17:10 Let me show you; I will click terminate and gone.

  • 00:17:12 I cannot recover back any of my data.

  • 00:17:15 So this is the major difference between using the community cloud or secure cloud without

  • 00:17:22 storage; storage only works with the secure cloud.

  • 00:17:26 And how you will run it again.

  • 00:17:28 So I will connect to the Jupyter lab one more time, then to re-start it:

  • 00:17:34 when you go to the very bottom of the instructions, you will see that I have this command.

  • 00:17:39 This will kill the automatically started web UI and start a new one.

  • 00:17:44 So let's copy this.

  • 00:17:45 Let's just wait for the Jupyter lab to load because we moved our data, so it will take

  • 00:17:51 some time for our data to be synchronizing with the new pod, new machine.

  • 00:17:58 So this is a shared storage that is between different machines.

  • 00:18:02 Okay, let's try again.

  • 00:18:04 Yeah, we need to wait a sometime, synchronization may take some time.

  • 00:18:08 Okay, let's try again.

  • 00:18:09 I think synchronization has been completed.

  • 00:18:12 You will also see the VRAM usage once it's loaded; it started.

  • 00:18:17 You can also wait here to load some memory.

  • 00:18:20 Let's connect to the Jupyter lab.

  • 00:18:22 Let's copy the command one more time.

  • 00:18:24 Yeah, Jupyter interface loaded.

  • 00:18:27 Let's open a terminal, copy paste the command, and the previously automatically started Web UI

  • 00:18:32 instance is killed.

  • 00:18:34 And now it is starting a new instance.

  • 00:18:36 The advantage of this technique is that you can see whatever is happening on the terminal.

  • 00:18:42 If you depend on the initially started web UI, you can't see anything.

  • 00:18:46 So this is a much better way of doing it.

  • 00:18:49 By the way, if you have noticed, I have removed the skip install argument from the web UI

  • 00:18:56 arguments.

  • 00:18:57 So with this way, you can also install other and newer extensions as well.

  • 00:19:03 Without removing that skip install argument, you won't be able to.

  • 00:19:07 Moreover, enable insecure extension access is added as an argument.

  • 00:19:12 So this also allows you to install any extension from the web UI interface itself.

  • 00:19:19 Once the Automatic1111 Web UI has started, you will see that it is running on a local

  • 00:19:24 URL like this.

  • 00:19:25 Don't click this; it won't work.

  • 00:19:27 You need to connect with the HTTP proxy port, and it will start.

  • 00:19:31 You can also alternatively edit the web UI user.sh file, which is here, web UI user.sh,

  • 00:19:39 and in here, you can add --share.

  • 00:19:40 This way, it will also give you a Gradio share live link that you can use anywhere,

  • 00:19:46 even on your phone or on another computer.

  • 00:19:50 And the interface started, and the rest is the same; you can just use it as you wish.

  • 00:19:54 So this is the way of using the network volume, how to use it.

  • 00:19:58 So I terminate and then I can deploy to any other pod.

  • 00:20:02 This applies to anything and everything.

  • 00:20:05 It is not only specific to Stable Diffusion web UI; you can use this technique on literally

  • 00:20:12 everything, on all of my other scripts or whatever you are running for.

  • 00:20:16 Alright, the installation of the secure cloud is completed, you see running on a local URL.

  • 00:20:23 So to connect it, we will use the proxy connection.

  • 00:20:28 Let's go to the secure cloud instance.

  • 00:20:30 And in here, you see it also started using GPU.

  • 00:20:34 So I will click the connect to HTTP port, just wait for it to load.

  • 00:20:39 And we got the interface.

  • 00:20:41 You see it selected Hyper-Realism; TensorRT is installed.

  • 00:20:45 We also have the ControlNet; let me show you the ControlNet is here.

  • 00:20:50 And when I click here, you see all of the models are automatically downloaded and ready

  • 00:20:56 to use.

  • 00:20:57 This is amazing

  • 00:20:58 believe me. This will save a lot of time of yours.

  • 00:21:00 So for example, let's try to use this.

  • 00:21:04 We can use this as a line art.

  • 00:21:07 Let's see.

  • 00:21:08 Okay, probably we don't need to process; but we can process.

  • 00:21:11 Okay, line art standard from white and black and white SD 1.5 selected is SD1.5 model.

  • 00:21:18 An amazing flying creature.

  • 00:21:21 Let's try 1024 1024 and hit generate.

  • 00:21:27 If you don't know how to use control that I have a huge tutorial the link is here, so

  • 00:21:30 you can watch and learn it, and we got the ControlNet output.

  • 00:21:35 This may not be the best setting because I don't remember; maybe we need to not select

  • 00:21:40 any preprocessor here.

  • 00:21:42 And let's see the output.

  • 00:21:45 Okay, yeah, let's try like this way.

  • 00:21:47 It should be pretty fast too.

  • 00:21:49 Yes, let's see the generated image.

  • 00:21:52 Okay, yeah, without preprocessor, it didn't work.

  • 00:21:54 So we need a preprocessor.

  • 00:21:56 So you can try different preprocessors and see which one is working best.

  • 00:22:02 And let's see the other one is, yes, it is also completed.

  • 00:22:06 So we can also start using the community cloud.

  • 00:22:08 Now, there is one important thing that I need to show you; recently, ControlNet added the

  • 00:22:17 InstantID and IP adapter face ID; using them is not straightforward.

  • 00:22:21 So I added the links of them to here.

  • 00:22:23 I suggest you watch this to learn how to use them.

  • 00:22:27 For example, to be able to use InstantID, you need to make the first control that unit

  • 00:22:33 like this and the second one like this.

  • 00:22:36 However, from my testing, it is not working very well.

  • 00:22:39 So let's select the SDXL version from here.

  • 00:22:43 And let's upload face here.

  • 00:22:46 Okay, let's use the Arnold Schwarzenegger, perhaps.

  • 00:22:49 Yes, this one.

  • 00:22:51 And from here, it says you to select InstantID face embedding and IP adapter InstantID.

  • 00:22:58 Okay, I will do that.

  • 00:23:00 So first of all, let's click the InstantID.

  • 00:23:03 And in here, let's select face embedding.

  • 00:23:06 And in here, IP adapter InstantID.

  • 00:23:07 So this was the first ControlNet unit; we have selected it.

  • 00:23:12 And these are default settings.

  • 00:23:15 Let's click the preview if it will show the preview.

  • 00:23:18 I don't know if the preview is working or not.

  • 00:23:20 With my scripts, you can also watch what is happening in this terminal because we are

  • 00:23:26 starting them from the terminal.

  • 00:23:27 By the way, you will also get this attribute error.

  • 00:23:30 This is the error of TensorRT; I think this will get fixed when they update the TensorRT.

  • 00:23:36 So the preview is getting.

  • 00:23:38 Yes, it is downloading the necessary models.

  • 00:23:40 Okay, we should get the preview in a moment.

  • 00:23:44 It downloaded the necessary models for the InsightFace library.

  • 00:23:49 And yes, we got the preprocessor output.

  • 00:23:52 I think it is not very important.

  • 00:23:53 So let's also select, for example, this pose, okay, this will be our pose, I think.

  • 00:23:59 So here we select IP adapter, and we select, okay, we select InstantID, sorry, key points.

  • 00:24:07 So this will be our face transferred into another one.

  • 00:24:11 And in here, we will select IP adapter InstantID SDXL, let's verify from here, ControlNet

  • 00:24:17 InstantID, InstantID face points, okay, so this will be face key points and ControlNet

  • 00:24:23 IstantID.

  • 00:24:24 Alright, and let's define just a man.

  • 00:24:27 Okay, and let's see what we will get.

  • 00:24:29 I will show you same with my own developer InstantID gradio and I am still developing

  • 00:24:35 it hopefully it will become even better.

  • 00:24:37 The link is here to install it.

  • 00:24:40 You can see one click automatic installer for the InstantID so I will just start it

  • 00:24:45 it is in my R drive inside.

  • 00:24:48 Let's see InstantID.

  • 00:24:49 Okay, let's start the web UI.

  • 00:24:52 Okay, let's go with high VRAM and everything.

  • 00:24:56 And let's see the result.

  • 00:24:57 Okay, so this is the result we got.

  • 00:24:59 Let's see if we have any errors.

  • 00:25:02 No, it doesn't show any error.

  • 00:25:04 And let's try again.

  • 00:25:06 So we enabled we selected key InstantID face key points InstantID control.

  • 00:25:12 Okay, it is enabled.

  • 00:25:14 And here we got the other one face embedding.

  • 00:25:17 IP adapter InstantID and the key points and control InstantID.

  • 00:25:23 This is the output we got.

  • 00:25:25 And in here, let's see if there is any error.

  • 00:25:28 Okay, I think we got an error.

  • 00:25:30 Okay, it says that yeah, it is it is still seeing another ControlNet it says. InstantID.

  • 00:25:37 No, it is looking accurate.

  • 00:25:39 InstantID let's try again.

  • 00:25:41 Okay, it think it taught that for some reason we have selected another ControlNet model

  • 00:25:46 but we didn't select so it is loading IP Adapter InstantID SDXL ControlNet model

  • 00:25:51 IP Adapter InstantID SDXL. Okay, everything is loaded.

  • 00:25:54 Meanwhile, my gradio also loaded.

  • 00:25:56 So let's load the same images and see what we will get.

  • 00:26:01 I'm just waiting for this one to be completed first.

  • 00:26:03 Yeah, it is loading everything accurately right now.

  • 00:26:06 Okay, ControlNet, using preprocessor, using InstantID face keypoints.

  • 00:26:10 Okay, generating images.

  • 00:26:12 This is not important.

  • 00:26:13 As I said, this is for the TensorRT error.

  • 00:26:17 I hope they fix it sooner.

  • 00:26:18 Okay, it is also pretty slow 1.76 seconds. In my Gradio

  • 00:26:24 let's try like this way.

  • 00:26:26 So I will transfer the face of Muhammad Ali, this is an upscaled by my upscaler.

  • 00:26:33 This is an upscaled image with the SUPIR which is a newest upscaler which is best I just

  • 00:26:40 made a tutorial for it.

  • 00:26:42 And let's use that face.

  • 00:26:44 And as a pose, let's use the Arnold Schwarzenegger this pose.

  • 00:26:48 Okay, so this will be our reference and this will be the face transfer.

  • 00:26:52 So let's say a man holding a sword and we have amazing features you can apply LoRAs

  • 00:26:59 you can select any models.

  • 00:27:01 Hopefully I will make a tutorial for this.

  • 00:27:03 So I will be by using the default model.

  • 00:27:07 And to get this pose, we can enable Canny or Depth.

  • 00:27:12 Let's try with Depth and submit.

  • 00:27:15 I hope my VRAM is sufficient.

  • 00:27:17 I didn't enable the low VRAM.

  • 00:27:19 And this is the output we got.

  • 00:27:21 Yeah.

  • 00:27:22 So I tried to make this I don't know this is working very, very bad.

  • 00:27:26 We are following the instructions they provided but it is not working good.

  • 00:27:30 Maybe let's try with control weight or maybe let's try with pixel perfect.

  • 00:27:34 So I make the pixel perfect.

  • 00:27:36 And is this also pixel perfect selected?

  • 00:27:39 Yes, let's try again.

  • 00:27:41 And let's see the result of my gradio I hope I don't get into shared VRAM because it will

  • 00:27:45 take too much time.

  • 00:27:46 By the way, this is another important thing.

  • 00:27:48 If you get shared VRAM as I am having right now, it will become 20 times slower.

  • 00:27:55 So you need to reduce your VRAM usage.

  • 00:27:58 Since I am recording right now.

  • 00:27:59 Okay, we may have an issue.

  • 00:28:02 Yes, yeah, my FPS has been dropped.

  • 00:28:05 This is bad.

  • 00:28:06 Okay, it turns out that I was also having open SUPIR upscaler.

  • 00:28:12 So it was using my extra VRAM.

  • 00:28:14 I closed it and shared VRAM has dropped.

  • 00:28:18 Now it started processing very fast.

  • 00:28:20 So if your GPU starts using shared VRAM, it will become 20 times or more slower.

  • 00:28:27 Be careful with that.

  • 00:28:29 We got the output of the ControlNet inside Automatic1111 Web UI with the InstantID and

  • 00:28:34 this is another output we got it is just horrific.

  • 00:28:37 So let's try to make the ControlNet weight like 0.5 and 0.5 and try again.

  • 00:28:44 And this is the output we got with the InstantID that we have.

  • 00:28:49 If you want to get a perfect match, then you need to do is upload same person different

  • 00:28:56 pose.

  • 00:28:57 So it will also match the face.

  • 00:28:58 What I mean is let me show you.

  • 00:29:00 So here my photo.

  • 00:29:02 Now, here my another photo, since the face will match, by the way, these are both DreamBooth

  • 00:29:08 generated images.

  • 00:29:10 Now I can generate much better matching pictures.

  • 00:29:14 So let's say a man standing on a chair.

  • 00:29:19 Okay, by the way, if we make this Canny it will not work very well.

  • 00:29:22 Let's make this also Depth and try again.

  • 00:29:25 Let's see this one.

  • 00:29:26 Yeah, this is still not working.

  • 00:29:27 I don't know how we can make this one working.

  • 00:29:30 I am following the instructions written on the ControlNet.

  • 00:29:33 This is official repository and this is the official thread where they have explained

  • 00:29:39 how to use InstantID.

  • 00:29:41 It is simply not working.

  • 00:29:44 And this is the output I have.

  • 00:29:45 You see, it is pretty accurate.

  • 00:29:47 It may even get more accurate depending on whatever you choose.

  • 00:29:51 So if you delete this option, it will use this also as a ControlNet.

  • 00:29:55 For example, you can use this as a pose and let's say a man wearing an armor. Currently

  • 00:30:03 the watercolor style is selected.

  • 00:30:06 So you can also try other ones.

  • 00:30:08 Let's try vibrant color and try to generate. You can play with it and get very accurate

  • 00:30:15 faces.

  • 00:30:16 Very accurate results.

  • 00:30:17 Let me show you one of the results that I have got.

  • 00:30:20 I have shown in the Patreon post, which is let me show you. Here.

  • 00:30:26 So you see, I have used GlowNeon XL V1 LoRA model and I have used SDXL base

  • 00:30:34 model.

  • 00:30:35 This was the input image. I set a man lightning.

  • 00:30:39 And this is the output I have got.

  • 00:30:41 I didn't select any ControlNet for this example.

  • 00:30:44 And it worked perfect.

  • 00:30:45 So you need to play with this to understand how it works.

  • 00:30:48 You see, this is also pretty accurate.

  • 00:30:50 But in here, we cannot get anything like it.

  • 00:30:53 I don't know why.

  • 00:30:54 Let's upload the same picture to demonstrate again.

  • 00:30:58 So I will use the same picture in both of the images.

  • 00:31:02 And let's say a man drawing in 3D CGI, or I don't know, we can say anything.

  • 00:31:11 Let's try again.

  • 00:31:12 It is just simply not working as supposed to be.

  • 00:31:15 I think it is also slower than my configuration.

  • 00:31:19 It is taking really high time.

  • 00:31:21 Alright, we can play with anything.

  • 00:31:23 We have some you see the styles here.

  • 00:31:26 For example, let's try neon and try again.

  • 00:31:30 And this also generating. Okay, this is the output we are getting and the ControlNet

  • 00:31:34 weight is just 0.5.

  • 00:31:35 When I make this one, it will become even worse.

  • 00:31:39 And we have selected face key points.

  • 00:31:42 InstantID SDXL, we selected face ID embedding, IP adapter, InstantID SDXL.

  • 00:31:47 We made everything true.

  • 00:31:49 And you see, it is like even it is not even these pictures.

  • 00:31:53 There is certainly some error someplace.

  • 00:31:57 Maybe it was using preview.

  • 00:31:58 I don't know.

  • 00:31:59 Both of the ControlNets are enabled.

  • 00:32:01 Let's try again.

  • 00:32:02 And here, this is working amazing.

  • 00:32:04 You can see. So this is everything for today.

  • 00:32:07 If you have enjoyed it, please like, subscribe to our channel, you can always contact me

  • 00:32:13 from Patreon.

  • 00:32:14 I keep everything updated.

  • 00:32:17 And let's just wait this one too.

  • 00:32:19 Okay, this time we got an image that is actually like similar.

  • 00:32:24 Yeah, somewhat similar.

  • 00:32:25 So we uploaded the same person.

  • 00:32:28 And we said a man drawing in 3D CGI, it tried drawing a man.

  • 00:32:32 And yeah, I can see some resemblance, but still nothing like the standalone gradio.

  • 00:32:36 Hopefully, see you in another amazing tutorial video.

  • 00:32:37 I will hopefully make a tutorial video about InstantID as well with better configuration

  • 00:32:38 and settings.

Clone this wiki locally