00:01Hello
00:03So, you clicked on the video, yes, but why did you click? You naughty, naughty, you teasing
00:05But seriously, it's good that you clicked because today I'm going to show you exactly how I made that thumbnail using AI, step by step, no tech tree required
00:16In this video, we're going to be generating these images and so many more, because we all need to learn how to use AI
00:24Whether you're an artist, a content creator, or just a curious human being, learning to use AI tools is essential
00:28So, if this is the way we are going to learn AI, then be it
00:32At least you'll know how stuff works and it's going to give you a Starting point to explore more ideas
00:37Now, I know what you're thinking, do I need to know Python, command lines, hack NASA? Nope
00:43All you need is a bit of curiosity and you're good to go
00:49In this video, I'm going to walk you through where to find image generation models
00:55There are so many of those, how to run them locally, and how to use ComfyUI
01:01ComfyUI is a tool that looks super complicated at start, but trust me, it's not that complicated
01:06By the end of this video, you'll be clicking buttons like a pro
01:13And yes, we'll also go over uncensored models for academic reasons, of course
01:17If you have a laptop With a GPU, good
01:20If you don't, then go to my channel and look for the last video that I posted where I show you how to set up your own AI server
01:24I'm going to be using that server for this video, but even if you want to run image generation on your own laptop, on your own hardware, the same instructions apply
01:37The only other thing that you'll need to do image generation is you need to smash that like button
01:40It won't work if you don't do it
01:43Trust me, I have tried that
01:46It doesn't work
01:49I am 100% telling the truth
01:52So, by the end of this video, you'll know how to install ComfyUI, load image generation models, and Start making AI art or thumbnails on your own system completely offline
02:05I'm going to hide my face now so that you can fully focus on the video, and with that being said, let's get right into it
02:08All right, people
02:12So, we are in the Google Cloud Console
02:17This is my project
02:22I've got a bunch of other VMs
02:27Uh, VM stands for virtual machine, but the one that we made last time was this AI tests, and as you can see, I have turned all of these uh VM instances off because I am not using them, and it's a good practice to turn them off because otherwise they're going to cost you money or credits
02:32So before I turn on this AI test, Let's see what this VM is about, what kind of CPU and memory and GPU it has
02:38So, let's see that
02:45Uh, if we click on that, we go down and we can see the machine type is N1 standard 2
02:51It's got two CPUs and 7 1/2 gigs of memory
02:58I think this is maybe a bit less, uh, it can still work, but it's good to add more juice to the machine, so we can, we will be doing that now
03:04Uh, it's got a GPU, one Nvidia T4, is good enough
03:10And then the last thing is the boot disc
03:14It's got a 50 gigs uh hard disk
03:18We can increase that too
03:22So, we go and click on edit and we go down, down, down, and in here you can either select a preset of machine type
03:28So, You've got these ones: four CPUs, 15 gigs, eight CPUs, 30 gigs, but the pricing remains the same whether you use a preset or a custom
03:36So I like to have custom so that I have more control over how many CPUs I need
03:45Let's say I need six CPUs, that's three cores, and memory is the thing that you need to have, uh, when doing AI stuff
03:54So for this one, let's go with 32 gigs of memory
04:01The reason I am not going with the preset is because, uh, you don't get 632 preset, so you need to do a custom one
04:06And, uh, then we go down to the boot disc, and, uh, I don't think they can update that here, so it's going to be in a different place, no
04:15worries, and the rest of the stuff stays the same, and we can just click on save
04:24So this is now updated, it's saved, and we can see now that the machine type is custom 6, 32, 768.6 is the CPU
04:30This is the memory, 32, 768, 32 gigs
04:36And let's update the boot disc as well
04:40We 50 is a bit too less, maybe, maybe we'll need more
04:45So we click on this one, it's going to take us to the disk section any moment now
04:48So we are here in manage disks
04:52Uh, this is, this is, this is our disk AI test
04:56It's being used by, uh, our VM that is AI test
05:00So we go click on edit, and in here you can increase the disk size
05:05So let's make it 100 gigs, and that's pretty
05:07much that we need to change
05:10Everything is okay
05:13Click on save, and it is saved
05:16Now, one thing to note is that this you need to uh do when the VM is turned off
05:20Some things you can change when the VM is turned on, but some things you cannot
05:24So, it's it's good that if uh you make the changes when it's turned off
05:28One important thing to know is that this is only for people who are going to use Google Cloud or any other cloud provider to use any VM
05:34If you do have your own hardware, your own laptop that is uh good enough, that is, that has enough memory, that has um enough GPU, then you can skip this part
05:46So now that we have Uh, made changes to our VM, we can turn it on
05:52It's, it's going to take a few seconds
05:58All right, the VM is turned on now
06:03Uh, the next thing we need to do is, uh, set the Windows password
06:10I don't remember what I had set the last time, so I'm just going to reset it
06:16So if you click on set Windows password and, uh, we reset the password for this username and click on set, it's going to give us a password that you need to save because it's only going to be for one time
06:22It's not going to be shown to you again
06:29And now that we have done that, we click again this and download the RTP file, and then we click on this one, and we click on Connect
06:48We enter the username and password, and we click okay, and we click yes, and it's going to start an RDP session for us
06:57This is all, uh, some unnecessary stuff, so you can just close everything here and here, and that's it
07:06Now, you've got a personal virtual machine
07:14To make sure that you've got the GPU drivers, you can just go to device manager and click on display adapters and see if you've got the GPU loaded onto the machine
07:22And in this case, we do have it
07:29If you don't, then you need, then in my previous video, I showed you how to do it, so, uh, that can be consulted
07:36So if you have your own laptop, This is the point where you should start following the instructions or the tutorial, because up until this point, this was, uh, just setting up or starting the virtual machine and, uh, getting into it
07:48So now that we are in the virtual machine, let's get into the juicy bits
07:59One thing to note, I, uh, added a new disk, or I add, uh, increased the disk size, but in here, you can see it's not added, and that's because we need to, uh, add the disk space also inside Windows
08:05And for that is, uh, create and format hard disk partitions
08:11You go here, and here you'll see that 50GB is unallocated, so you just right click this, extend
08:18Volume, and next, next, finish
08:24And now you've got, uh, now you've allocated the unused disk space
08:31So you go here, you click refresh, and now you will see that you've got about 100 gigs of disk space
08:36So let's get into the juicy bits then
08:40And the first thing we need to do is install ComfyUI
08:44So if you go to any browser, Edge is the default that comes with Windows
08:50I hate this
08:55And in here, if we can search ComfyUI install, blocks the first link should be okay
09:01I prefer the portable version, it's much better
09:07So we don't need to do any installation
09:14You go here and, uh, just click this one and it's a sevenzip
09:17file
09:21It's 2 gigs, but the good thing is that we are in a data center, so look at the speed here
09:24Holy, yeah
09:28So 2 gigs was downloaded in a matter of seconds
09:31Now it's going to do the scanning part
09:34That's, that's Windows doing its thing
09:38Uh, it's going to take some time
09:41All right, so this got downloaded, and at the same time, I also downloaded 7-Zip because we needed that to, uh, extract that file
09:44So if we go to our downloads, and this is a 7-Zip file, I just say go that and extract to this
09:51It's going to take some time because it's two gigs
09:58So it got, uh, unzipped into this directory, this folder, which is, uh, 6.2
10:02gigs now
10:07So double click, double click again, and this is the thing that we needed
10:11So, uh, since we have a GPU installed, we will be running this file called "run Nvidia GPU," and that's pretty much it
10:18We double click this thing, it's going to do something, and let's see how it goes
10:24It's going to take some time
10:29Maybe the first time you run it, it's going to take some time because it's loading all the stuff, but the second time it's going to be very fast
10:35Here you can see it's got it, it found the total RAM that was 32 gigs
10:42The VR RAM is 15 gigs, but we can increase it if we, if, if there's a need, but let's see if, if There's a need for that or not, and here it goes
10:50It's going to open up the UI automatically
10:56The main thing to note here is that this is running on the local server, so this is not on the internet
11:02This is completely on your own machine, and uh, it's very private
11:09So the fastest way to start using this ComfyUI is to use these pre-built templates
11:15So there is uh, image generation, there is image to image, there is this Lora, and there is lots of stuff
11:23So let's start with uh, basic image generation, or there's also flux
11:27Flux are, is this is really, really good model
11:32So let's start with maybe flux dev
11:36It's uh, Pretty interesting
11:40Or on second thoughts, just let's start with basics, uh, basic image generation, and then we can move on to the flex to see what that one does
11:44I keep changing my mind, I'm sorry
11:48Let's do image generation, click on that, and it's going to say that
11:51Okay, it's, uh, let's close this for now
11:55So, what you see here is a bunch of boxes, that's all I can describe it as
11:58It looks scary, but it's not that scary
12:04So let me explain to you some basic things
12:11First of all, let's change the save image into a preview image because I don't want to save images every time I generate an image, I want to
12:17Preview it first, and then if I want, then I can save it
12:22So I click on delete this trash button
12:26I click, double click, and when I do double click, there is this preview image
12:30You can also search for it, but it's just right there
12:35So I can just place, uh, say preview image
12:41Uh, this is a note, and then I just click on this and drag it, drag it, drag it, and then drop it here, and that's it
12:46One thing that you can, you can, you can also do is to, uh, resize this thing so that you can at least see what kind of image you are generating
12:53Then the most important thing here is this block, this load checkpoint, and this thing is where you load
12:59Your model
13:05A model is something that generates the image for you, and there is a website where you can go download models
13:11There are so many of them, but we will be finding maybe just one and downloading that, and using that one
13:19We can save this workflow since I made some changes
13:27Let's just say image generation
13:32The place to look for models is a website called Civit AI
13:38I have it opened in here, and if you go take a look at the homepage, you are greeted with this
13:43Do note that you need to have an account to access the, uh, this to access the models
13:49So I have already logged in, but one thing to make Sure, is that your NSFW filters are turned on, because there is a lot of NSFW content in here
13:57Uh, if you want to avoid that, then I suggest that you turn it on
14:06It's turned on, uh, by default for my account, so I hopefully won't be seeing anything
14:12There is some stuff that is still passing the filters
14:19Well, it's not passing it because, uh, there is no NSFW content in here, but as you can see from the name, it, it suggests that it is for that stuff
14:23But we are good people, so we will not be doing that
14:27Instead, let's go over the basic concepts of what this thing is
14:33So we are looking for something called a checkpoint
14:39So as This uh node says we have to load a checkpoint, so there's lots of stuff here
14:45There is, well, this is a checkpoint, this is a checkpoint, this is a checkpoint, but if you scroll down, uh, there is, uh, also something called Allora
14:51I did not want to click that
14:57I can go back
15:04Uh, I'm not going into the details of what Allora is
15:10Um, there is also some other, there's like a workflows, uh, and other stuff, so, uh, for now, we will be only focusing on the checkpoints, so we can sort out, uh, the models
15:14Let's, this is being sorted by most downloaded
15:19You can do that, or you can just go with the highest rated, and it's going to
15:24Show you what has been, uh, what people have been using or downloading the most, and, uh, for some reason, uh, this one is being downloaded the most, but then there is this, this, and this one
15:41So for this one, we can either go with any of the models in here, or I have been experimenting with a model called Epic Realism, and the reason I was experimenting with this was that this is the model that came up when I was browsing through the most downloaded the last time
15:55So now it's not showing up, but for some reason, this was the most downloaded the last time, and I gave it a try, and I liked it, so this was a good one
16:00You can
16:05See the pictures that have been generated by this model
16:09The name is Epic Realism Excel, and then you have to just search it in here, and it's going to bring you to this one
16:14The idea here is to just learn how to use the models rather than get fixated on one model, so this is just an example
16:20You can use the same steps to use any other model
16:26So here are the pictures that are generated using this model
16:32These look really good ones, so I would like to give this one a try
16:41So what we do is that, uh, we, uh, so this is the latest version, the realism one, and we just click on download, uh, and it's going to start downloading
16:53So the thing is that this is, we have a really, we have high internet speed because we are sitting inside a data center, that like, this virtual machine is sitting inside a data center
16:57Like, look at this speed
17:01This is insane
17:05So six and a half gigs is going to download in under a minute
17:08So while this is being done, I'm going to show you a few other stuff
17:11So the reviews, you can look at the reviews of the model
17:13This one has is overwhelmingly positive
17:16The base model is SDXL
17:19There are other base models as well
17:23There is Pony, and then there is Stable Diffusion, other models, but the Excel one is usually, uh, good to use normally
17:27What happens is that when you open a model, the creators of the model, they have shared instructions on how to use it
17:38You don't see for this one in here for some reason, but if you go back and take a look at any other model, I think so there's going to be instructions
17:45So let's go take a look at a random uh model
17:50So Cyber Realistic XL, let's take a look at that
17:56So here you can see the instructions, the uh recommended settings, what they have uh used, and the example prompts that they have used, and some other stuff
18:04So you, you can use all this stuff, but in our case, I think our model has been downloaded, so it's
18:09This one, uh, what we need to do now is, so this is going to be in our downloads folder
18:17So, uh, as you can see, I downloaded this already one time, but let's use this one
18:21So what we need to do is that, uh, I can just delete this one
18:26It's six and a half gigs, so I can just delete this, the old one
18:30This one, what we need to do is we need to do Ctrl X, and, uh, we need to put it into the right folder so that ComfyUI can read it
18:36So what that right folder is, it's basically you go inside, uh, ComfyUI folder, then you go to models, then you go to checkpoints, and here you can see it says, "put checkpoints here," and we just paste that thing here, and now the, uh, ComfyUI will be able to read it
18:51It's not going to read it by default, so you would need to restart ComfyUI
19:07So I'm just going to close this thing, and I'm going to close this thing as well, and I'm going to go back to the start page and just run this thing again
19:22So when you click on this one, it's going to find now the model that we have put in, in the checkpoints folder
19:28So you just click that, and now the model is in there
19:33So now the example that we want to generate, you can be, uh, you can be artistic and write whatever you want to write here, and it's going to try to generate that thing, but for this model, I would, or, or For this tutorial, what I would like to do is, uh, use the examples that they have provided, and the name of that model was, I already forgot it, the name was Epic Realism
20:00So this one
20:07So if you click on any of these pictures, it's going to show you the prompt that has been used to generate that picture
20:10So this is the prompt that has been used
20:14Uh, if, and we can just directly copy that prompt
20:18So copy that, and we go here, and now here are two boxes that you might be thinking what they are
20:21So this one is the positive prompt
20:29So this is something that we would like to generate, and, uh, that is the prompt that you find in here, this box
20:35However, is uh, the stuff that you don't want to generate, and that is called a negative prompt
20:41So in this one, what we don't want is that it generates a cartoon, CGI illustration, a painting
20:47So for this particular case, we want that it does not generate this stuff
20:54And then you can see some other stuff too
21:00So there is CFG scale that they have set for this picture, that was three, and that thing that you set, that thing in the K sampler node, and the CFG is this one
21:06Now if you hover over this one, it's going to show you what this means
21:12It says that it's the scale balances creativity and adherence to the prompt
21:18Higher Values result in images closely matching to the prompt, but too high values will, of course, negatively impact
21:24So, in simple terms, if the value is higher, then it's going to match the score, uh, the prompt as much as it can
21:30If that value is lower, it's going to, uh, be more creative
21:37So these people, they have, uh, put the CFG scale at three
21:43I would say, uh, I would like to keep it at four or something
21:47Then there is something called steps
21:52Steps is the, uh, value you can set that thing in here
21:55Steps is how many times you want the pictures to be refined into, uh, the picture that you want
21:59So what happens is that this latent Image, this is just a blank image
22:03That blank image is generated with the seed noise, and with every step, that image, that latent image, is refined into, or it's converted into, the thing that we want to generate
22:12So, in step one, there's some pixels that are removed
22:20In step two, some more pixels are removed, and then in steps 20, 25, you get the final picture
22:29Some models are really good at refining or generating the picture in a few steps, like three or four steps
22:34Uh, some models require more steps
22:40The more the number of steps, the more time it takes to generate the picture, uh, but of course, then the picture is more Sharp, more clear, and is, is, uh, generally good
22:46So they have used 28 steps
22:52I would say, okay, we can, for this first demo, we can also use 28 steps, and we can do enter
22:56Then there is the sampler
23:01Sampler is, I will not go into the details, but if you click on this one, you can see there's lots of samplers in here
23:05Uh, I don't know what, like, I, I can't find the exact sampler name that is mentioned here
23:11In here, the latest, the closest that you can find is this one, DPMPBP2M or, uh, SDGPU, but I found that the default of Oiler, it just, uh, works fine, and then it's just going to decode it, and then we, we will be able to preview the
23:21image
23:30So if I just click run, I think there's going to be an error, but let's see what error that is
23:35Yes, and I was expecting that error
23:40Uh, this is uh an Nvidia related error, so you might also encounter that
23:45If you don't, that's good, but if you do, what you need to do is just like simply copy this thing and just go here and paste it
23:52And I think there was uh a GitHub issue somewhere
23:59Maybe it's this one where they said, so if you go to this one and you read through it, uh, you will find that you need to add something called, you need to add this thing, disable Koda Malik, in uh, in the startup script
24:12So you copy that
24:16And then you go to the script
24:20Uh, this is the start, the startup script that we have
24:26Well, first of all, let's close this thing
24:32Uh, you right-click this, you click edit, you go to the very end, and then you paste that thing, and you do a, since this is not saved, you do a Ctrl+S, and then you stop it, and then you run this thing again
24:37So now that it's started up again, uh, we can see we can just run that thing again
24:40We can close this, and that's the old one
24:43We can close that one too
24:46So now that is uh done, we can just simply run this thing
24:49All the settings are the same, so if I click on run, the first run, it's going to be a bit slow
24:55Because it's uh loading the model, but this was pretty fast, but and now you can see it's it's gone there, and I'm not even going to speed up the video
25:06It's it's going to be in real time
25:18So, the K sampler is doing its thing, and it's almost done
25:23And there you have it
25:28This is very similar to the one we saw in the example picture
25:33So, look at the the details in this one, is this is very very uh realistic
25:38And if you want to compare it, uh I think I accidentally closed it, so I can just go back to that epic realism, epic realism XL
25:50So this was the picture that uh we generated, and this is the picture that we got
25:55So it's very Very similar, I would say
26:00So it's very, very similar
26:05And if you want to save this image, then you just right click and click on save image, and it's going to be saved
26:09But let's go and take a look at some other examples
26:14So let's see if you want to generate this one
26:18We copy this prompt, we go here, and we just say paste
26:25And this one is like a house bar, perched attention, and a rock
26:32If for some reason you don't know what to put in here, you just have an idea of what you want to do, go to Chat GPT and, uh, just ask it to generate a prompt for you, and it's going to, uh, do its thing
26:38So this one doesn't have a negative prompt, so we can Exclude a negative prompt in this one, and now if I click on run, it's going to be much faster than last time, even though the last time was also fast
26:51But wait a second, did I click? So this is 3.532
26:54Uh, I think 28 is good enough steps and four
26:57I'm happy with cfg of four
27:01You can experiment with this stuff
27:04Uh, you can experiment with the different results that you get
27:07You can, uh, experiment with the sampler, uh, type
27:10You get different results based on, uh, what sampler you are using
27:16But if I click on run now, as you can see, the model did not load because this was already loaded and it's just doing its thing, which is Super, super fast
27:22And again, I'm not, I'm not going to speed up the video, it's happening in real time
27:27There are some models that take a lot of time, but this one seems to be pretty fast, and we are done now
27:42Okay, this is very good quality, very, very good quality, I'm, I'm surprised
27:48And this is the image that was uh generated as an example, and we got the exact same thing just in, just as a mirror image
27:53So uh we can do one more example
27:58For example, let's uh let's, let's do this one, let's see what do we get
28:03Copy that thing, go here, uh control V
28:08This is this is a bit uh detailed prompt, so it's portrait masterpiece
28:134K blah blah blah and the negative prompt
28:18There is some negative prompt, which is going to be like bad proportions, ugly, bad hands, bad teeth, and stuff
28:24So let's just click on run and see what happens, and maybe I'm going to skip, uh, speed up this one, uh, just to give you the results
28:31Now, this one is, um, not that great in my opinion
28:36This one is much better
28:40Uh, this one is same face, but not, not that great
28:44Uh, so if I click on run again, it's going to just start, uh, the node from here
28:50It's not going to reload all these, uh, nodes that did not change, and we get an image
28:55Well, this is much more similar to this one, except for The eyes, they're not that great
29:01Uh, we can try to increase, uh, or decrease
29:05Let's, let's see if I make it go to eight
29:09Let's see what happens
29:12All right
29:16This is, uh, I would say better than, than the last time, but you can just play around with the, with, with the settings
29:19So, uh, for example, this one, I'm not so sure how that's going to turn out
29:24We can try this one
29:28Let's see
29:32So I go here, Ctrl A, Ctrl V, and in this one, this doesn't have any negative prompt, so delete that
29:36I have a suspicion that it's, it's going to, it's going to mess up the pictures in the background, so let's see what happens
29:43Well, all was similar
29:49But yeah, as I suspected, this did, it did mess up the pictures in the background, so that is not such, not such a good example, but yeah, you get the idea
29:55Okay, let's do this last one, uh, from these example pictures, and this was the prompt
30:04There is a negative prompt, so I go here and I just do that thing, and let's keep the CFG back to four and generate or run
30:21All right, this is actually a really good photo, so this one and this one, they match very much, like look at the quality of this one
30:28So that's how you generate the uh pictures
30:31That's how you gen uh use these models
30:35As I said, there are lots of models, uh, lots of
30:39Different kinds of models
30:42There is, uh, video generation models as well
30:46There is lots of other embeddings, uh, here
30:50Embeddings are, uh, well, something for another video
30:55But for now, I think this is, uh, this is good enough, uh, for a starting point
31:01So you can just go and download these models and play around and and be creative and learn, uh, how to use this, uh, ComfyUI and and to use these models
31:07So have fun, play around, learn stuff, and most importantly, don't be evil, elves
31:14Until next time, see you.
Este guia ensina como usar o ComfyUI para gerar imagens com modelos de IA, mesmo sem experiência técnica, cobrindo desde a configuração do ambiente até a geração de imagens. 
Esta seção detalha a instalação do ComfyUI e os primeiros passos para gerar imagens, incluindo a navegação na interface e a preparação para carregar modelos. 
