Revealed: This Is Exactly How I Make My Thumbnails
LARRUZ walks through their thumbnail process: generate a glitchy background in Midjourney, create a posed likeness in Stable Diffusion using DreamBooth and ControlNet, upscale in Canva, then add logo and effects. The whole workflow uses AI tools and takes about an hour.
Chapters
I'm going to show you exactly how I make my YouTube thumbnails. This is probably the most requested video that I've had in the comment section of my YouTube channel. And I'm going to dive in and show you right now. So this is actually for a video that I recently made about why Bing Chat is actually so bad. So the image that I have in mind for this thumbnail is a bright, colorful image, just like I do with most of my thumbnails, but with maybe like a cracking or glitching out Bing logo. And then my face kind of giving like a. Like a this isn't good kind of look. That's sort of the image that I have for this. Now, when I make these thumbnails, it all starts in Mid Journey. I like to make the background of the thumbnail using Mid Journey because Mid Journey just has these vibrant colors that just really pop. Now, if you don't have a Mid Journey account already, you can go and get a free trial of it over@midjourney.com and with that free trial, you can generate up to 25 images, I believe. And then after that, their premium account starts at like 10 bucks a month. Now, in order to use Mid Journey, you actually have to also have a Discord account. Discord is completely free. You can Download it@discord.com I have several videos that explain Discord and Mid Journey. So if you're not familiar with Mid Journey, maybe go and start with one of those videos first. Now, once you're in Mid Journey, if you're still on the free trial account, you'd log into Mid Journey's Discord here and you'd come into one of these newbie rooms and you would generate your images in one of these newbie rooms. So since I'm on one of the paid accounts, I actually get to use Midjourney in a private room and I don't have to deal with everybody else generating their images. Now because the theme of this video is Bing sort of glitching out, I'm going to try to make a colorful background image that represents a glitch. And if you've watched my previous Mid Journey videos, you would know that I like to use less in my prompts with Mid Journey because the less I give it, the more kind of creative Midjourney gets with the prompts. And I usually end up generating images that are just better than what I would have come up with with with my own brain. So I'm going to let Mid Journey do a little bit of the work. For me to get the background image. So let's go ahead and type Imagine. And since this is referencing Microsoft Bing, I like to throw the brand name in there because sometimes it will try to incorporate that brand's logo into the image. So let's put Microsoft, and then let's put glitch. Let's add RGB here. When you add rgb, it stands for red, green, blue. It usually makes it so there's a lot of color in the imagery. And then I like to add this little rainbow emoji because then that really makes sure there's a lot of color in the image. Let's go ahead and put no people. I don't want it to have any people in the imagery. And let's go ahead and do Our aspect ratio 16.9. Because a YouTube thumbnail, the standard aspect ratio is 16.9. So Microsoft glitch, RGB, rainbow emoji, no people, aspect ratio 16.9. Let's see what that gives us. Okay, so it totally ignored my no people prompt here. So let's go ahead and put Imagine. I'm going to go ahead and copy this same prompt up here and paste this one down here. So Microsoft glitch, rgb, no people, no face face. Let's see if we could get it to ignore faces as well. Okay, so I actually rerolled it a couple times here until I got some images that I liked. So here was the first round that it came up with. These didn't really represent what I was going for. This one kind of did with a computer sort of melting down here. But I don't know, it looks a little like too dark and evil to me. But then it came up with this one, which I really like because it's kind of got that Microsoft logo going on there, and it's a computer falling apart. And that could kind of be representative of the Bing Sydney situation breaking down a little bit. So I'm going to roll with this image and see how it comes together when I go to fully compose everything. So I'll go ahead and upscale number four here. And while we're waiting for that to upscale, I'm going to jump into stable diffusion. Now, I prefer to use run diffusion in the cloud because quite honestly, I can run a faster GPU in the cloud on run diffusion than I would be running on my own computer right now. Now, if you've never watched this video on my channel here called Inject yourself into the AI and make any image with your face, that explains the process of how I actually got my face into stable diffusion so that I can make images generated with my face. The rest of this video is probably not going to make sense to you unless you've gone through that whole process of using Dream Booth and training your likeness into stable diffusion. All right, so once you've actually done that and you've got stable diffusion installed, which I have a video on that as well, called Install stable Diffusion locally Quick setup guide that will show you how to set up stable diffusion, and it also explains how to use run diffusion to use stable diffusion in the cloud. So now that you've got stable diffusion set up, or you're running it in the cloud and you've trained your face into a stable diffusion model, you should be able to follow along to the rest of this video. So in order to get the exact pose that I want, I'm going to use the new control net extension, which you can add to stable diffusion. If you're using run diffusion, it's already pre installed. So all you got to do is have a run diffusion account and it'll work for you. So I want an image of disappointment, just like a. Ugh, my head in my hand like that. So what I'm going to do is I'm going to go to Google Images and see if I can find an image that's similar to the pose that I want to have in the image that I'm going to generate. So let's go to images.google.com and let's do Disappointed Face. Let's see what it finds for us. So an image like this would be perfect. I want to make sure I have the proper licenses to use it. It's sort of a gray area right now of whether or not I can use an image like this. Even though I'm not using this image itself, I would be sort of training stable diffusion to give me a pose like this. And so I don't really know how that whole thing plays out. So I'm gonna go ahead and go to a stock photo site, which I have some credits at, and try to find a similar image here. Let's go. Face palm. So I found this image on a stock photo site. I'm gonna go ahead and purchase it. That way I know I'm not gonna be getting in any trouble for using this image for my pose. Again, I'm not gonna actually use this image in my thumbnail. I'm just using it to get the pose from it. All right, so I'm gonna jump back into stable diffusion here, and I'm gonna open up control Net down here and I'm image that I want to use the pose for. And I'm just going to drag it and drop it right in here to make sure I use this pose. I'm going to make sure I enable Control Net for my pre processor. I'm going to use depth and then select depth again here. All right, so I entered a prompt here and just to give you some context, Dominic Cooper is actually a famous actor. But it's also the prompt that I trained myself on back when I trained my likeness into this model the first time around. I had heard that if you give it a name similar to someone that you kind of look like, then it would work a little bit better. I don't actually believe that's true anymore, but back when I made this model, that's how I did it. So Dominic Cooper is the keyword that I use in my prompts to make the image look like me. So I'm going to do a photo of Dominic Cooper. Person, Face, Palm Upset Digital painting artstation Concept Art Smooth, sharp focus. Illustration Art by Art Germ and Donato Giancolo and Joseph Christian Leyendecker Ross Tran Wlop so these are all keywords that I've gotten good success with to get my image from in the past. Some of the negative keywords I like to use ugly, disfigured, weird, cut off, extra fingers, deformed, extra limbs, cut off, cropped, no head, fat, chubby, puffy, weird teeth. I also like to put suit and tie because a lot of times it likes to generate me in a suit for some reason. For sampling method, I like to use DDIM sampling steps. I'm going to go to 40 here. I'm going to leave this width and height at 512. I want the batch size to be 4. I want it to give me 4 images. I'm going to let the seed be random. So if this is a negative one, it'll be a random seed here. CFG scale. I'm going to leave it 12. And then we've got our control net down here set to depth. And when I generate this, it's going to look kind of like me, but you're going to see it's not going to look exactly like me. But I'm going to fix that in the next step. So let's go ahead and generate these four images and see how they come out. And here's what it came up with. There's the first image, there's the second, there's the third, and there's the Fourth, now you can see they kind of look like me a little bit, but not totally. I think I'm going to go ahead and work with this one right here. And so what I'm going to do is I'm going to send this one to InPaint. I'm going to click on Send InPaint, and you can see it regenerated it over here in inpaint. And I want to sort of paint where my face is and try to get it to look a little bit closer to what I look like. So I'm just going to carefully try to trace around the hair and the face, like so. Don't need to get it all because it's already kind of close. Ish. The beard's not quite there, so I want to get rid of the beard because it's made it a little bit too long. And so I'm just going to go ahead and paint that out like so. Up here, I'm going to put parentheses around Dominic Cooper and person so it knows that's kind of what to put the focus on here. And then down under my inpaint settings. I want to make sure this is selected on inpaint masked. And then in paint area, I want it to be only masked. Ideally, we get some images that look more like my face in here. So let's go ahead and change our batch size to four. Give it four shots at it. Bring our CFG scale up a little bit here, and let's go and generate four of these, and let's see what it gives us. All right, so these are the images that it generated here. You can see it kind of looks closer to my face. So let's kind of take a peek. That one's a little off still. That one gave me a neck beard, that one gave me a really long beard, and this one actually looks pretty spot on. This is kind of what I'm going for here. I mean, this hair is a little bit darker than my actual hair color, but I'll work with it. I think that's close enough. I think people will see the thumbnail and get the idea that that's supposed to be me. So let's go ahead and make sure we save this one. And now it's time to pull it all together. So to pull it all together, I use Canva. If I come over here to canva.com, come up to the top right and click Create a Design. There's an option to create a YouTube thumbnail, which already sets it at the perfect size for a YouTube thumbnail. So I'll go ahead and click this. Let's jump back into mid journey. It should be done upscaling by now. I will open this in a new browser, save our upscaled version, and let's go ahead and start composing our image here. So let me open my downloads folder here. So I've got this open. Let's go ahead and pull in our background image that we created here. And let's go ahead and pull in this image of my face palm here. All right, so we've got these. So I'm going to go ahead and right click on this and set images the background. So now we've got this colorful kind of Windows computer melting down. I kind of want it to be shifted to the left a little bit here. So I'm going to go ahead and select my background, come up to crop, and then I need to make it a little bit larger in order to be able to shift it around. And let's try to get it offset to the left a little bit here. Now we've got this sort of computer breaking down here. And then we've got me doing a face palm. Now if you're on the paid Canva plan, you can actually come up to Edit image and click Background Remover and it will actually take out the background of the image here. If you don't have the paid Canva plan, you're using a free version of Canva. There's tons of websites where you can remove the background. In fact, if you go to futuretools IO and you go type in Remove background. This one right here, this REM BG is a free background remover that you can use. But since I do have a paid Canva account, I just do it right here in Canva and you can see it remove the background. So now I've got my likeness trained into the image. I'll make it sort of this larger image here and then just kind of position it the way I like it. Now this background's looking a little grainy to me, so I actually want to upscale it a bit. So one more time I'm going to come over to Future Tools here. I'm going to type upscale this UPSC A Y L. This is a free upscaler here, which is really good. There's this one called AI Image Upscaler, which I really like and it's free and you can do it right in the browser. So I'm going to go ahead and use this one real quick. So let's grab this background image that we Created. Just drop it in here. Going to take a second to upscale it. I'm going to go ahead and click on Enhance Quality. And then I'm going to go ahead and click on Download Image. Now, I have an upscaled version while I'm in here. Might as well upscale the image that I made of my face as well. So let's go ahead and pull this one in as well, and we'll get an upscale of that one just so both of them are a little bit higher quality. Go ahead and click on Enhance Quality again. Download this image here. All right, now I've got better versions of both of these images, so I'm going to go ahead and delete both of these here. Pull in this upscaled background. That image is my background. And then pull in this upscaled version of my face here. And once again, I'm going to go ahead and remove the background on this one. Go ahead and select our background again and bring it up a little bit larger. We upscaled it so we can do that now, and we didn't lose much quality there. I could go ahead and pull this up a lot larger now, too, because it's been upscaled. And I'll go ahead and put that right there. And then I want the Bing logo in here in some way because this is all about Bing. So what I'm going to do is I'm going to just do a quick Google search here and search for. For Bing, and we'll go ahead and download the Bing logo real quick. I'll go ahead and pull the Bing logo in. So now we've got that inside of our image here. But I want it to look kind of glitchy. So what I'll do is I'll click on Edit Image. And inside of Canva, they actually have like a little digital glitch kind of feature here. They call it Slice. So I could kind of see what these various glitch effects look like. And I think I like. I think I like this minced one because you can still tell that it's the logo. So I'll go ahead and pull this one in, bring that up a little bit, and it doesn't really pop off the background much. So I'm going to go ahead and add a shadow behind it. I like to use the glow. Go ahead and make it a dark glow. Turn the transparency up so it's a lot more dark. Turn this blur up so it spreads it out a little bit more. And then let's adjust the size like that so it really kind of Pops out of the image. Kind of like the idea of maybe it being sort of in the monitor or hovering above the monitor. I want to make this background even bigger. I'm going to zoom out a little bit, make it even bigger so that this monitor is more towards the left of the screen here. All the colors in the monitor anyway, so I'm not losing any color on this. And then I'm going to move this so it's sort of like in the middle of the monitor like that. Move myself a little over, and I'm going to give myself a little bit of a glow, too, so that I stand off the background a little bit. So I'm going to click on shadows, add a little glow here on myself. Use our settings here. Bring up the transparency so it stands out a little bit more. Bring up the blur a little bit and bring out the size slightly. So now I kind of pop off the background a little bit, too. And then I just want to add my text to the image. So I will go ahead and click on text here, add a heading. And then I sort of have a style that I like to use with all of my images. I click on Effects, and then I just put this background on it. And then I make the text itself white, and I make the background down here black. And then I give it a slight transparency so you can see through it a little bit. And that's kind of my style for my thumbnails. It's. I don't know, it's just something I always do on my thumbnails because I just like the look of it. And then I'll put. Let's go, Bing is broken. Just do something like that. I like it to be a little bit of a tease. I don't like to say exactly what's going on in the video inside of my thumbnail. And then I'm going to sort of center it with where this monitor is here, too. So I'm just kind of trying to get that centered a little bit. Then I could do something like that. And there we go. This is how I make my thumbnails. I've got my face trained into the image. I've got a real colorful background. I've got the Bing logo sort of glitching out. And then I've got my text on the screen that sort of teases at what this video is going to be about. And there you have it. There's the whole entire process of how I make my thumbnails for YouTube. So now you can stop asking for a quick recap I go into midjourney and I create a colorful background in mid journey. I then go into stable diffusion because I've already trained my face using Dream Booth into stable diffusion. And I generate an image using a pose that I find online so I can get the exact pose that I'm looking for. And then I pull it all together in Canva and add some text and, you know, maybe some extra elements like the Bing logo into it. That's the whole process. I wouldn't say it's the fastest process in the world. There's definitely a lot faster ways to make YouTube thumbnails. But this is the process that I use to get these really cool images that you see when you look at my YouTube channel. You know, 90% of this thumbnail was generated with AI. So for someone like me who has just never been good at art, I can't draw a stick figure. You've seen in some of my previous videos when I've tried to draw a cat or a penguin, I can't draw to save my life, yet I could make images that look like this. That is what I love about AI. That's what makes AI so amazing to me, is that people that never really had the ability to create artwork that looks really good now are enabled to be creative and to take the vision that they have in their mind and turn it into something. That is why I think AI is so cool now. I know a lot of people are out there going, that's not art. You didn't create that. But it's my vision turned into an image. I had this thought in my mind of what I wanted to look like. I wanted a glitching out computer with the Bing logo. I wanted a face palm image of myself. And I was able to take that thought that I had into my mind and make it into an image that looks like this. That is what makes this stuff so powerful. And if you're as excited and passionate about this AI stuff as I am, come nerd out with me over on FutureTools IO. This is where I save all of the cool tools that I come across in the AI space and I curate them and I organize them for you so that you should be able to find them more easily. And if there's too many tools, you don't want to learn about all these tools. Make sure you just sign up for the free weekly newsletter every single Friday. I send you the five coolest tools that I came across throughout the week. I also send you a handful of news articles, a handful of YouTube videos, and one way to make money with AI so that you kind of get the TLDR of the AI space for the week. I send it every Friday and it's free. You just got to head over to Future Tools IO. So thank you so much for tuning into this video. I hope you enjoyed nerding out with me. If you like this video, give it a thumbs up down below. And if you haven't already, subscribe to the channel. Because I'm going to keep on making new tutorials and I'm going to keep on nerding out and showing you all of the latest and greatest AI tech on this channel. And hopefully I'm going to continue to up level my videos as well so that you find it more enjoyable and more help. So click that, subscribe and that like and I really appreciate it. Thanks again for tuning in. See you guys in the next one. Bye.
