How to Create Stunning AI Thumbnails for YouTube

Happy 2026. Hey, how's this for a new Year's resolution? How about you master AI marketing? The AI Business Society gives you live training with experts who are getting real results. No hype, no fluff, just proven workflows that work. And you've got access to an incredible library of training from many of the incredible guests that you've heard on this show. So secure your AI transformation by visiting socialmediaexaminer.com AI welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzer, and this is the podcast for marketers, creators and business owners who want to know how to use AI. I'm really excited about today's interview. I'm going to be joined by Mike Russell and we're going to explore how to create thumbnails for YouTube videos with artificial intelligence. Now, whether or not you are a graphic designer or not, and especially if you're not, you're going to find this absolutely amazing because the stuff you can do with AI is going to blow your mind and you're going to want to stick around to the end of the interview to hear about face swapping. Yes, you heard that right, Face swapping. By the way, I'm at Stelzner on Facebook and ikestelsner on X Twitter, and if you're new to this podcast, be sure to follow this show so you don't miss any of our future content. Let's transition over to this week's interview with Mike Russell, helping you simplify your AI journey. Here is this week's expert guide. Today, I am very excited to be joined by Mike Russell. Russell, if you don't know who Mike is, you've got to know Mike. He is the founder of an amazing company called Music Radio Creative, which is an audio agency for creators. And he also has an AI focused YouTube channel called Creator Magic. And he's got a podcast by the same name, Creator Magic Mike. Welcome to the show. I'm super excited to be interviewing you today.

Super cool to be here. I can't wait. There's so much to talk about.

Today. Mike and I are going to explore how to create amazing looking YouTube thumbnails utilizing AI tools. Now, before we go there, Mike, we gotta hear a bit of your backstory. Like, how'd you get into AI? How'd you get into audio? Start wherever you want to start, because I know there's a bit of a story there. I can't wait to hear it.

Yeah, it's exciting for sure. So I started out years ago working in the radio industry, so I guess I was an original audio content creator. My main interest was putting together amazing sounding jingles and pieces of audio. But I've always had a passion for tech, so technology has always excited me. In fact, I've taught on YouTube, YouTube for over 10 years on my original channel. And the thing that excites me the most about teaching on YouTube is talking about new features and new ideas. And obviously I've always been excited by new Adobe releases and things like that, but now with AI, there's like a release every day, so it like really keeps me on my toes and it's kind of going back to my roots because since I was a kid I was tinkering with back in the day it was a BBC microcomputer and the old Spectrum you may remember, and then in the later years it was the Raspberry PI's and stuff. So I've always had an interest in tech, so AI is, is logical for me.

Well, and let's share a little bit more of the story. Music Radio Creator, for those that do not know, is the company that creates all the sounds that you hear for all four of our podcasts and even some of the other shows we've done in the past. So all those awesome intros and the transitions, they do all of that. They also edit all of our shows. Like tell us a little bit more about Music Radio Creative and then tell me how you got down the rabbit hole of AI.

Sure. Thank you. So Music Radio Creative, as you rightly mentioned, is a creative audio agency. We create intros and outros for podcasters. We do full production and editing. We also work with people making video podcasts as well, and we help you to sound great. So whether you've got a radio show or you're a DJ or you're a business looking for audio, or if in my case you simply want a bit of coaching on how to set up a microphone and get a room sounding good without too much echo going on or removing background noise, I can still pretty much teach that now. But I think AI is fast coming in and ChatGPT is giving the answers to a lot of the questions that I've answered and coached people on over the years. So hence why, I guess, yeah, we can move this way. Maybe. I hit the existential Crisis when the ChatGPT moment dropped and I was like, my goodness, I'm typing in every question that I've got about, like, Adobe Audition, the audio editing software, and it's answering it in like 10 seconds flat better than I could do. And I was like, what is my value in the world? I don't know what I'm going to do with this. So I. I hit this moment where I was like, this is replacing me fast. And I know we're probably going to get deeper down this rabbit hole. A lot of what we do in our business, we work with voiceover artists now. AI is so human sounding. There are tools like Suno and Udio that can create music. That's something we've been doing for years and now it's gone up to a standard where it is incred. Incredible. So I think it's not just me. A lot of creators and businesses and marketers out there would have been concerned and maybe still are concerned by AI.

So why did you decide to start a YouTube channel focused on AI when you knew it was potentially a threat?

That is such a great question. I'm excited by the technology. So I'm a firm believer that AI is here to help augment us and enhance our skills and make us better creators, better marketers, better business people. And the subject we're going to talk about in this show today, today is thumbnail design. You may know that I'm not a thumbnail designer. I am an experienced audio producer, audio editor. I know a lot about audio, but I've always wanted to do the full thing. And AI has really helped me dial in my thumbnails in an incredible way. So that's why I want to share about that.

Well, and what kind of stuff are you talking about on your channel? I know it's more than just what we're going to be talking about today. Sure.

So Creator Magic is a channel where we explore AI tools, specifically with a focus for content creation. So if you're making a podcast, if you're creating a YouTube channel, if you're going live and you want the best AI tools to do the job, I help you to discover them. And I do it weekly on that channel by surfacing the best tools that I think are around that particular week. As you mentioned at the start, I have a podcast where I interview other content creators who are using tools and I figure out how they're using AI and their takes on AI and really very much new. It's only a few months old. It's my playground to play with and discover new tools and find the best way to use AI in life in general.

Folks, I know that Mike is world class. And he's not going to say it, but I'm going to say it. So Mike is the guy, the leading guy in the world when it comes to audio. Like so many of my friends who are audio snobs, as I call them, people like Cliff Ravenscraft, who are very particular with their audio, look up to people like Mike Russell, because Mike understands everything about audio. He knows Adobe Audition like the back of his hand, which is why his crew not just does the music jingles, but all the editing for all of our shows. And he always has, and they're amazing. And what's great about Mike is he is very technical, but he's also very creative. And that's what I love about today, is that Mike is one of those creators like me, who has gone into kind of this new frontier that we're calling artificial intelligence. And he wants to help people understand what can be possible with it. So I'm just going to state that from the outset that Mike is a world class creator who's new to the world of AI in the same way that I am. And I'm really excited to have him on the show today. And you know, Mike, you brought up some of the existential crisis challenges that people have been facing. You know, creators, marketers, entrepreneurs that run agencies and stuff. We're all feeling the stress. And my question to you is, what do you want to say to people that are creating out there that are resistant to embracing artificial intelligence? Because obviously you have overcome it. Help them overcome it.

Well, thank you so much for saying that. And I think the short answer is embrace AI or you will be replaced. You'll be replaced by someone who's using AI. We've heard many experts in the field saying similar phrases, but it really is true. Because imagine you have one narrow skill like I do, which is the audio world. And then imagine AI allows you to branch out and say, okay, I'd like to know a little bit about script writing, thumbnail design. Now, it's not suddenly going to make us experts or understand the nuance, but it certainly is going to get us closer to making incredible content. And the way I see it right now, and of course it's developing all the time, but I see it as the best second employee or intern with a wide base of knowledge. And the better I use, the better I prompt at the moment, the better the responses I get back. So if I'm looking for help with a title for my YouTube video or I Want some ideas on the thumbnail? I can start to ask some ideas. And I use it as really, I think copilot is a great word. It's like a copilot as I sit here in my office working away, really super intelligent. And it can help me be more productive and more creative, I think, than I've ever been in my life. And I'd encourage anyone listening right now who's worried about their future to really embrace, sit and learn it and try and cut out the overwhelm. Because I know that many people, certainly I feel this way, feel there's so much going on, there are so many releases, there are so many tools. Which one do I use? And really, it can be very simple. And I think, as you mentioned, Mike, that's one thing that I try to do is take complex problems and break them down into very simple answers and solutions. And I do believe that AI will help with that.

So my appeal to anyone who is creative that's listening right now is just give this podcast a listen and take it to the end, because I think that you're going to find that there is a way that I can help you in a way that you're not strong. Mike's not talking about necessarily using AI on the audio side of things, although he could be. He's talking about using AI in areas where he is not strong, where he's becoming stronger. And as a result of this, he's able to do things he could never do before, or he's able to do things that he would have had to pay a lot of money to someone else to do. And now he can do himself for practically nothing but his time. And that's where I think the opportunity is for us who are creators is that we can now create more or we can create better, or we can have get it done in less time and have more time to do the things that we love, like go to the beach or travel or hang out with our family. So there. There's so many cool things that are here. So let's start with this first question about what do we need to be thinking about? How ought we reframe our mind when it comes to starting to think about utilizing AI from a creat perspective?

So when you start to think about using AI from a creator's perspective, I think you need to go on that journey from the very start. So it's, what piece of content do I want to create? And start to ideate yourself. So I always start with my own mind thinking of, like, what is it that I want to do. I'll make notes. Usually I'll use Notion to put together some ideas and then you can go over to somewhere like ChatGPT and ask it some questions. What do you think about this? How can I make sure I don't leave anything out when I talk about this subject? You know, what are the questions? What are the problems? What are the pains that people are experiencing on this particular subject? And I can start to get a list and refine it and make it ready for me. And then I think you really need to find your go to tools that you're going to use and stick with them. And actually, I would say despite the buzz that happens in the AI space, don't try to keep up. You know, follow people who can help you keep up, but find the tools that really work for you and then learn how to use them. And I know later on we'll talk about should use Midjourney for image generation. Should I use Dall E for image generation? And they both have their strengths and weaknesses, but find those tools that really work for you, learn how to use them, and then go ahead and try things out.

Well, I want to come back to something you said that you just barely touched on that I think is really important, which is ask the tool like Chat GPT. What do you think? Expand on that a little bit. What does that mean? Because not Everybody understands that ChatGPT gives.

You a response that is as good as the question you ask it. So you can have an idea, say, for a YouTube video, and you can say, right, this is what I want to talk about. And then you can go ahead and say, I'm thinking of talking about this subject. So let me give you an example. For instance, I'm thinking of talking about making a video on thumbnail design. What should I make sure is included in this video? And ChatGPT will respond with a bunch of ideas. It'll say, well, you need to think about which editor you're going to use to edit the graphics, which AI image generator you might use, and what steps you may take to make that. So you're really working as a team to find out what it is you need to include in the video. So, yeah, I mean, there's so much you can do and it all really depends on what you ask, I suppose.

Well, and what I like about where you're going with this example is you could say, hey, I'm going to make a video about hiring an editor, a graphic designer to create YouTube thumbnails. And I'm believing that I want to cover these things. What is their experience in my niche? Let's just, you know, freeform a little bit. What is their primary graphic style and what are the tools that they're using to create the graphics? Right. This is what I'm thinking about doing with my video. You could input that into chat GPT and you could say, what do you think? And by the way, here, here's who my target audience is. Right. Do you feel like this is a good theme, this is a good direction and then it's going to tell you. Right. Well, maybe you might want to consider this or maybe you should emphasize this and de. Emphasize this. This is kind of where we're going with giving it your plans and asking it for feedback because now it becomes a consultant for you. Right. And that's where sometimes we don't have anyone else to talk to. And rather than asking it for ideas, you could ask it for what does it think about your ideas? Is that kind of what I'm hearing you say?

Absolutely, yeah. And it's brilliant. And while it can help you write a full script for a video, I love to ask for an output where it will give me bullet points. So I use a prompter in front of me, an elgato prompter, and I will put those bullet points in front of me on the prompter as I present the video. So I can make sure that I'm not leaving anything out and I can go through logically. And the best thing is it doesn't stop there. So I can get it to help me plan, then I can deliver the video. I can then transcribe the video I just recorded, feed it back to ChatGPT and say, how did I do? Did I get everything right? Is there anything I missed? And then based on that, you know, do you think this is a good title for the video? And let's start to talk about thumbnail ideas as well. So I mean, you can use it all the way along the journey.

Love it. Okay, so speaking of thumbnail ideas, let's get into this. Okay. You have a very cool process and we're gonna go deep into it right now. Where do you start with thumbnails? And just in case anybody listening right now doesn't even understand, you know, we're talking about YouTube thumbnails, right? And typically we're talking about something that's a 16 by 9 design and we're mostly talking about long form YouTube video. We're not talking about shorts here. So just, just to clarify, I mean, that's my Understanding what we're talking about when we're talking about thumbnails, right?

Yes, absolutely.

So where do we start?

We definitely started already with the ideation inside ChatGPT. You know, give me the, the topic, the talking points, but then further to, you can ask ChatGPT. Okay, so what would be a good thumbnail for this? And it can suggest some ideas. So say we're talking about thumbnail design. It might say someone sitting there painting a 16 by 9 image on a canvas or something like that. And you can think, okay, that's cool. So that could be, you know, a picture of me with a paintbrush in hand painting on a canvas or an electronic screen or something like that, and you start to get the idea. Other times I'll just get an idea in my mind as I'm ready to start designing the thumbnail and I'll think, it has to look like this. So I did a video fairly recently about an AI tool that brings Claude and ChatGPT and lots of other AI chats together in one place. And I thought it has to be a big superhero with a cape wearing the logos all around. And then you can go ahead and start prompting to create those images. And that's where again, you might use ChatGPT to come up with prompt ideas as to what needs to be in the image. And that's where the fun starts. But I don't know how fast you want me to go through each step. So we're kind of coming up with the idea.

Let's zoom in a little bit on coming up with ideas. Let's just assume not everybody is as creative as you are, Mike, and they, they're not going to come up with a guy with a cape on and a bunch of logos on the cape. Right. Just because they don't have Mike GPT like you and I have. Right? So, so let's start with like the basics. Like how do you actually ask a tool like ChatGPT to come up with ideas for a thumbnail? Like, let's break it down, you know what I mean? Like, you've already made the video, presumably, and you know what the topic of the video is about. How in the world do you get good thumbnail ideas from ChatGPT? Let's start there, just ask it.

I mean, my primary way, once I've recorded the content, and I always think it's good going in before you even start to have a title and vague thumbnail idea in your mind before you even start recording so that you know that the video you're Making is representing the idea you have in your mind. But that's not to say once you finish recording, you can't feed the transcript back and say, take a look at everything I've spoken about, give me some thumbnail ideas. Give me an idea for an image that could really represent this long form video on YouTube and then it'll start coming back with ideas and you can tell it what direction you want to go. So initially it might be very generic. Like if you're talking about starting a podcast, it might say a big picture of a microphone. And you're like, there are so many other YouTube videos with a big picture of a microphone talking about podcasting. So you can say, okay, let's be more creative here. I want you to put a person in and I want you to have them doing something. I find for me, something that works for me is having some kind of loose theme around your channel. So for instance, on my channel, my loose theme is magic and wizardry and that kind of stuff. So I can always say, you know, a wizard casting a spell or holding the hands in the air or something like that. So then I can think, okay, if it was to be a video about podcasting, for instance, it could be a wizard, you know, shooting a lightning bolt at a microphone and sparks flying everywhere. But then, yeah, again, have a kind of loose theme I would recommend, and then ask ChatGPT, come up with some ideas. And sometimes it can just be really creative. It can give you three or four different ideas. And the more ideas you ask for, you can say, give me 10 different ideas, because it's likely that eight of them will probably be rubbish. And then there are two that will come out that you're like, oh, I could maybe go with that. Or go with that. It's a real idea spark.

I love a couple things you talked about. First of all, you said, hey, take the transcript and feed it in and ask it for recommendations on a thumbnail. That's an innovative idea. For those that don't have the transcript, you might have an outline that you created before you made the video. You could feed that outline in there and you can say, come up with recommendations for a good YouTube thumbnail theme. Right? And it's going to come up with a bunch of different ideas, right? And I would imagine you could use one of those ideas. Maybe you like one of them and you like the direction of it. And that, that's just that little spark maybe that we need to get in the right direction. I love coming up with abstract stuff. So doing the Thumbnails for this particular channel, I try to find some sort of a theme that seems to be the subject of the video. And then I typically ask it, you know, what's cool about ChatGPT and is that eventually it learns your style. You know, the more you do it, it'll come up with videos that are like, within your style. Okay, cool. So we've got some themes at this point and now we're at the point where do we want to talk about the actual prompts for the images or is there something else we want to talk about? Before we get into that, I'm just curious what your thoughts are.

Yeah, I think maybe just before we get into that, there is a tool that comes to mind that I've been using the last couple of weeks, and that is View Stats Pro. It's actually created by MrBeast and it's got an incredible feature in there called thumbnail search. So this is a paid upgrade. There is a free version of Vue Stats, but in the paid version you can type in a thumbnail idea. So recently I typed in something like someone meditating like that, and it gave me all the thumbnails across YouTube that have performed incredibly well for that subject. And I think it uses some kind of AI or machine learning image recognition to surface those thumbnails. So if you have a rough idea of what you want to do and you input that into this tool, it can give you the most successful thumbnails. And it shows you that by showing that this video has been an outlier on this channel or has, has had a significant amount of views and it's likely due to the thumbnail getting a decent click through rate. And it's incredible because it'll immediately show you a bunch of thumbnails on the screen and you can dive deeper down the rabbit hole with it. So you can say, show me more thumbnails like this and it will show you more thumbnails like that. And actually something I've been doing recently, literally in the last couple of weeks, is taking a thumbnail that I like the idea of. So either screenshotting it it or downloading it, and then I will feed it into GPT4 vision. So the latest ChatGPT vision where I can upload an image and I'll say, tell me what you see in this thumbnail. Describe everything as if you were describing it to an AI that could generate images from your text prompt. So already we're almost skipping over a step of having to come up with just text to describe it, because we can get a successful thumbnail, then use the vision to identify what's in that thumbnail. And then we've got a text prompt that we can feed back into dall e in the case of OpenAI, or over into somewhere like midjourney, which is another image generation tool. And I'll tell you, I did that and I got an incredible click through rate when I actually used that whole process end to end. And I was like, ooh, okay, I need to use this tool more.

Okay, so just a couple questions. View stats pro is Mr. Beast's software and is it like $25 a month or something like that? How much is it for the pro? Do you know off the top of your head?

I think it's around 49 bucks a month. Yeah.

Oh, 49. Okay. And that' the one that'll give you what you just talked about, which is you type in a certain kind of search query and it will show you high performing thumbnails. You can download that thumbnail and you can upload it into Chat GPT. You refer to Chat GPT Vision, but people may not understand. You can just use ChatGPT on your phone and take a picture, or you can use ChatGPT on the desktop and drag an image into it. Or is there a special version of it that you're using?

Exactly right, right. Yep, that's exactly what you do. Just download it, upload it into the chat, off you go.

Let's talk about filling the gap between AI potential and AI results. I'm sure you've tried ChatGPT. Are you seeing real ROI? Most marketers are barely scratching the surface of what's possible with AI. And it's because they're missing the training and the frameworks that will allow them to be very successful in their careers and in their future. AI business World allows you to get that training. When you attend, you get two full days and 20 expert practitioners, many of the people that you've heard on this very show, that will teach you exactly what you need to know to build the workflows and to succeed as an AI enhanced marketer. It takes place April 29th and 30th in Anaheim, right alongside Social Media Marketing World. And here's the good news. You can save $300 when you register by January 16th. Head to AIbusinessWorld live right now. This discount ends soon. What I love about it is you're asking it to describe it in words and explain why that is so important. Because now we're kind of getting into the prompting side of things.

So you can either start with an image like that or you can start with an idea. So say, you know, I've done a video before about saving time editing your audio, and it's a picture of me on a beach with like, you know, a nice drink sitting there, relaxing. So again, you'd say, okay, I want to make a thumbnail with me relaxing on a beach, drink in hand, no worries kind of thing. Describe this as you would to an AI image generator and then it will come up with lots of words that can be fed in for something like midjourney. It tends to like comma, separated words. I mean, it's getting more natural language adherent. I know we'll probably talk about the differences between Dall E and midjourney. So Dall E is very good at following a text based prompt, whereas midjourney tends to lead a little bit more guidance. But I find in my experience that the images you get back from midjourney tend to be far more photorealistic and creative, Dall? E a little less so. So if you were to describe, I'm on the beach with two drinks in hand and I'm wearing sandals and I don't know, there's a pineapple in front of me, Dali would likely do exactly that for you perfectly. Whereas you might feed the same prompt into midjourney and it might find you've only got one drink, or you've got three drinks in your hand and it hasn't quite cop. So there are pros and cons of each service. But get that text prompt from ChatGPT, bring it over to your image generator and start playing. And if you don't like things, you can go back to ChatGPT and say, could you tweak this prompt slightly and then bring it back over to mid journey or Dall E or whatever it is you're using.

Crazy question. Can you ask chatgpt to mid Journey this version? Mid journey is it. Do you understand what I'm asking? Can you ask ChatGPT to develop a prompt that is going to be acceptable for me midjourney? I just getting creative with you on the fly here.

You absolutely could. Now I remember in the early, early days when I was trying this out, it was not very good at doing that. And again, the reason for this is because of the training data, as you know. So if the training data was not up to date, it might not even have known in the early days what midjourney was. So it'd be like, well, here's a rough prompt and you're like, no, that's rubbish. It's clearly, I can see it's not going to work. So there are a few ways around that maybe now, now, I haven't tried that recently. You could ask for that and it probably would work pretty well. But another thing you may want to do if you're thinking of going down that avenue is searching online for Midjourney's best guidelines for prompting midjourney. I'm sure they probably have an FAQ where they say, best way to prompt is like this, this, and this. And put these tags on the end of your prompt if you want to get a 16 by 9 image, and so on. So what I would do in that case is I would copy that as long as the instructions are not too long. And I'd put them into Chat GPT as what is known as a custom instruction. So it's where you actually give a set of instructions that don't go in the chat box, but you actually say, okay, you are an expert AI image prompt generator. Please follow these guidelines. And then I would just paste in what I got back from midjourney's FAQ you. And then in the text box, I'd say, now make me an image of, you know, someone on a beach sipping a drink with no worries kind of thing. And then it will follow exactly those rules.

I love that. Okay, folks, this is a big tip right here. This is really big. I want everybody to, like, make sure they understand what Mike just talked about, because this applies beyond what we're talking about today. You can give Chat GPT a data set of instructions and then you can ask it to work, work off of that data set. That's a huge thing. That's an unlock I haven't heard before. That's so freaking cool. All right, so we're going to talk about mid journey versus Dall E3 in just a minute. But just let's talk general ideation when it comes to prompting for images. Like, do you have any tips on in general? Because most people just, they don't think the way you think. So, like, let's lay it out a little bit. Like when you are asking an AI image generator to create an image, what are some of the things we need to be thinking about?

Okay, so let's get on to what you need to think about. I also wanted to say on that set of custom instructions and prompting and creating that now I believe everyone, even free users of ChatGPT have access to GPTs, which are basically customized versions of ChatGPT. If you can create your own GPT with those custom instructions. It could be like, for instance, Mike's midjourney creator. You can put them there and every time just go and use that GPT and it'll know what you need. But yeah, as you just asked, you know, what am I trying to do? What am I trying to put there? So you want to set the scene, you want to say, you know, where is this going to be? Is it going to be a beach? Is it going to be in a park? Is it going to be, I don't know, on a boat or whatever? So you want to describe the broad scene, you want to get as much detail in there as possible. So you want to talk about what can you see. If you're talking about your mind's eye, what can you see in this image? Use as many words as you can to describe bit. And for my thumbnails, I always try to put a person in there. Now it's interesting because I'm sure in this episode we can get into the should it be a real person or not a real person? So on my original channel, my Mike Russell channel, it's the same name as me. I've used my real photo and my real image. But on my new channel, Creator Magic, I use my AI generated face. And that's quite an exciting process. And I think you can do more if you use the latter. So, so what I ask for when I'm asking something like midjourney to generate me a thumbnail, I will often make sure I say, give me a close up of the face or head and shoulders like that. Because sometimes if I say something, I'll be quite loose. To start with, it might be man standing on a beach holding a drink like that. And I get like a man really far away with a drink like that, standing with a palm tree next to. I'm like, no, no, no, no, no, I can't see that person. It's like close up, head and shoulders, holding a drink and making some expression. And the best thing is, again, again, we can get into this. Once you've got your thumbnail that you're happy with, you can do things like face swapping. And then if there's one little bit you're not happy with, you can either use something like inpainting in midjourney, where you can select a piece of the thumbnail and say, no, change this. So say you didn't like the shirt you had on. It was a big red shirt and you wanted a Hawaiian shirt. You could paint over it and Say, make a Hawaiian shirt or what I do. When I'm ready to really nail down my thumbnails, I'll take it into, like, Adobe Photoshop or anyone listening who does want to go use Photoshop because they're like, whoa, that's so many buttons and widgets. You can use Adobe Express, you know, it's. It's really easy.

Awesome. Okay, so here's what I heard you say. First of all, set the scene. Where is this actually taking place? If you don't set the scene, it's going to pick one for you that's really important for people to understand and then get as much detail as possible. That's also really something where I think the ChatGPT can help you set the scene. If you can't think of a scene, right, you can ask Chat GPT like, hey, I want to do an outdoor setting, but I don't know what to do. And here's what the topic is. Give me a couple of ideas so can help you narrow down the scene. Right. And then what should be in the screen, the. The set, if you will. I would imagine if you can't come up with creative ideas, you can ask Chat GPT to come up with those as well. Right. And I love the idea of putting a face in there. The final video for this will have Mike Russell's actual human face. It will not be an AI generated version for us, but we definitely agree with you on that. Now let's talk Dall E3 versus Midjourney Journey. Let's get into it. What are the pros and cons? I mean, let's just. You kind of hinted at some of it, but you love Mid Journey. I've never used Mid Journey. I'm very familiar with Dall E3. So let's just kind of start with what. What is it that Dall E3 cannot do, that Mid Journey can do? That's probably the best place to start.

So Mid Journey can do, in my opinion, far more photorealistic and incredibly detailed images. Dall? E, I think, think still extremely, extremely good, but not on the level. I think Midjourney just wears the crown. It is and has been the best AI image generator. That, of course, may change in the future. Some of the pros, though, of Dall? E, I would say, is it's baked into the ChatGPT experience. So you can do it all in one prompt. You can say, okay, I've got this idea for a video. What do you think about it? It'll say, yeah, this is good. Okay, I've recorded it. Here's my transcript. What are the thumbnail ideas? And it will give you five ideas. And you can say straight away to ChatGPT. I love idea number four. Let's see. Potential thumbnail. And immediately ChatGPT will make it using Dall E. So it's very, very easy. Also, as I said earlier, I think that Dall E is generally more prompt adherent. So if you want something to look a specific way, Dall E is probably going to do it a little bit better than midjourney. But I just think with midjourney, you just to have, have the best quality possible. Plus, in terms of doing styles and characters, as we may talk about, if you want an AI version of you, I think the character possibilities inside midjourney are really, really good.

One of the things that I am a little frustrated with with Dall E3 is a lot of the abstract stuff it does is just wonky. We have to end up pulling it into Photoshop and using Generative Fill to kind of like erase random stuff that looks like it's Russian, you know, it's not even English, or, or just weird stuff. You know what I mean? Does Mid Journey do an okay job with abstract concepts or is it more photorealistic? Like, let's say I wanted to create a maze, you know, and I wanted the maze to be a functional maze. Like would Mid Journey be able to accomplish something like that? Or is that more of a Dolly3 thing? I'm just curious what your thoughts are.

Ah, that's really interesting. I think both could do do an interesting job.

Because it sounds like you're using Mid Journey to create actual sets that look real in real life, right? Where sometimes let's say that I just want to create an abstract concept. So let's say that the concept is like productivity, right? And I just, you know, if I just ask Dolly to do something, it's going to. Well, I'll take a better one, like a balance. It'll end up coming up with a scale and it will show things that represent balance, you know what I mean? And it will show. It'll come up with a bunch of little images that kind of work that abstractly represent the concept of having balance. And that's the kind of stuff that I typically do. Is it true that Mid Journey is mostly kind of like what you would expect if you were to take a professional photograph? Is that kind of the difference between the function like everything looks legitimately real. You don't see these random things that make no sense when you zoom in on them. Or does that still happen in Mid journey as well?

Yeah, you can definitely still get lots of issues like that, like things that look out of place or. I mean, it's true. Back in the early days, we were so amazed by, say, Mid Journey generations. We were like, this is so realistic. But now I've heard many people say I can tell when an image is AI generated. And I think that's the risk we run of creating thumbnails or images that look like they're AI generated. And I think in some potential circumstances, that can be seen as a negative, because especially if you're targeting creative people, they can say, are they using AI to do their thumbnails and the images in their blog posts or whatever? And that's not good kind of thing. So almost like voiceover artists saying they're using AI to make this not a real human kind of thing. So I'll share. From my experience, one of the things I've noticed is if you try to really design a scene and lay it out and use ChatGPT to make the prompts, you can avoid things such as I found AI likes to be repetitive and create symmetrical kind of patterns and things like that. And yeah, like. Like abstract. So you're absolutely right, Mike, because if you say something like, you know, make me an image of scales, then it's going to make scales, but then it's going to maybe put, like, things floating off into the distance and things like that. But if you say, make me an image of a man standing on one side of the scales, the other side is flying up into the air and feathers are puffing into dust in the corner like that. And it's in a room that looks like this. And so the more specific you can be, the less chance you have of it going off abstract into kind of the fantasy, kind of almost. I call it like the wormhole world, where it just like repeats into infinity.

Well, and the other thing that I've also found is you have to tell Dall E3 that it's gotta be a 16 by 9 format too, because this is another tip. A lot of it just comes out square unless you tell it it's 16 by 9. And even when you tell it at 16 by 9, sometimes it'll create a 16 by 9 image with the square image inside of it, you know, and it's like. And it'll just kind of use random stuff to kind of. To kind of fill the rest of it. And I have to have my designer, I have to bring these images and then I have to zoom in on them and I have to tell my graphic designer. All right, here's where I think we should go with this. I think we should use Adobe Generative Fill to remove some of these things and zoom in on this particular section and frame it around this. And you know, it's. It's still quite a bit of work Mid Journey. My understanding with Mid Journey is it's far more technical. Is that fair to say that you have like, I've seen a lot of these prompts where you are calling out like the camera lens and all that kind of stuff. I mean, is that necessary or is that just. Just the function of the potential of this software, which seems more engineering focused?

Yeah. So I think you can certainly get like photographic styles if you call the lens and the style of image you want. I don't tend to go too far down that rabbit hole. So the best thing, like you say Dall E versus Midjourney. With Dall E, it might generate a square image and you say, no, make it 16 by nine and then it does it, but maybe not perfectly. Whereas with Midjourney you have things called tags. So you can put ar, which stands for aspect ratio, and then 16:9. So you know you're going to get a 16 by 9 thumbnail. You can put things like dash, dash, c, r, e, f and then a URL, which is a URL to an image you uploaded of yourself. And then it will try and create a character that looks similar to you. Now it's not exactly you. And Midjourney can only draw on its training data to create someone similar to you. So that's cref, which is character reference, will look at a photo you've uploaded and match most of your characteristics. But then you can go on and face swap later on. So yes, with Midjourney you have a lot more control by putting little flags or tags at the end to say, I want this aspect ratio, I want this kind of person in there.

Can you easily edit? You mentioned that Midjourney has. And by the way, ChatGPT has this. A lot of people don't realize it. Also, where you can, you can select an area and ask it to regenerate a section of it, but it's not great. You know, I think Adobe Generative feels better, but can you do that in Mid Journey? Is it easy to edit an image? Once you get an image in Mid Journey, can you say, hey, take that same image and change the exposure? Or I mean, tell me a little bit about the editing side of things.

Yeah, so that specific example, changing the exposure may be tough or the brightness. Yeah. What you could do is you could inpaint just the same as you you rightly mentioned, you can do in dall e. You can select over an area and say, turn this cocktail into a pipe or whatever. Then it will swap that out perfectly for you. Really incredibly well, actually. So inpainting is good if you've got a specific detailed thing you want to change in the image. But I'm absolutely with you. Like, I would get pretty much to where I'm happy with the image and then I'd pull it into somewhere like Photoshop. And this is the place where I can use generative fill to do small tweaks. And also they introduced just recently in an update, an exposure brush. So you can go if the face, you think, oh, it's not lit up very well, you can take this brush and paint over the face and brighten up the face really nicely. So that works incredibly well. Yeah, that's great. And there's so much that you can do in Photoshop. For instance, you know, I try to always take a leaf out of MrBeast's book. He tests his thumbnails. Like some of them he's like split testing them 70 different times to find out which one is the best. And he'll do like small things like change his facial expression and change the clothes he's wearing. So for instance, in one thumbnail it'll be a red shirt, the next one it'll be a white shirt, the next one it'll be a black shirt. And he goes as detailed as that. Like just changing small things. And generative fill is brilliant for that. Cause you can highlight the shirt and change it to different colors. You can change one element of the thumbnail. And I'm having a ball doing this because I just recently got access to the native a b testing inside YouTube studio and I'm having an incredible time finding out what works with thumbnails at the moment for my particular channel. It may not be the same for everyone listening right now who's testing it, you know, in a, in a different area. One thing for instance I found really works for me is having a very small amount of text, maybe two words on the thumbnail that explain what you can expect in the video. Because as well as YouTube looking at click through rate, actually my understanding is they're far more concerned with watch time. And I'm sure you'll agree with this as well, Mike. So they want the thumbnail to accurately match up with what the viewer is expecting to get in the video. So if you do that incredibly well and you get a big bigger watch time, they are going to make that the winner. One place I found this particularly exciting and interesting is with podcasts. So I have a podcast on the channel that I've just started. It's about four or five episodes in now. And I start off split testing with a picture of the person and then I'll put a little quote about what they said, something powerful they said in the show, like that. And what I can now do is I can have that picture with a little quote and I can test it three times. And what is brilliant is after the show's been out for a week or so, I can go back into YouTube studio and I can look at the retention on the podcast and I can see where the graph goes up and I'm like, oh, people were interested in that. Or where the graph goes static. They were staying and they were listening to that specific thing. I will then find out what was said there. This is, this is deep stuff. Now I'll take the transcript of what was said when the audience attention was high and I'll copy it, paste it into ChatGPT and I'll say turn this into a power powerful quote of no more than eight words. And then I'll use that quote on a new split tested thumbnail. So I'll find two or three points where I retained attention, copy those two or three points, make three powerful quotes and split test those powerful quotes until I really dial the thumbnail in to like what the podcast is about and what gets the most watch time.

Brilliant. And marketers listening right now are like, hey, I could also do that to create some other images to promote the YouTube video. All right, face transfer transplant. I can't not let you not talk about what the heck a face transplant is. Because when you said face transplant a couple times, people are like, what the heck is that? So give us the lowdown, what's the tools you use? And just be clear, what the heck you even mean? My face transplant tools, you know, Anyone.

Listening right now to totally understand, I'm not talking about the Nicholas Cage film. This is all done using AI and it is pretty much the final piece of the creation. I've got this thumbnail, thumbnail. The person in it looks kind of like me because I've used Midjourney's character reference, but it's not me. So I now use a tool to just swap my face onto that. And there are many ways you can do it from easy to hard. So one of the easiest ways that I'd recommend, and this is the way I'm doing it right now, is go to a site, it's called replicate.com and on there they have all kinds of AI tools. But if you go to that site and type in Face Swap, you'll find a bunch of different, different free face swaps. And all you need to do to use this site is have a GitHub account, which I know sounds techie, but it's free to have a GitHub account. So have one of those. Log in using GitHub and you can use their face swap tools. It's brilliant. It's free. You only need to upload one image of yourself and the thumbnail that you want to do the face swap on. Click Go. And the tool will do the face swap for you. So you can do it with replicate.com. there's also something like Insight Face, which is available in Discord. So if you're using Midjourney, you can add it, as I think it's known as a plugin or extension or something. So insightface will do that for Midjourney. If you're working inside Discord, if you wanna get really techy, you can install something like Face Fusion on your own server. But again, that requires a little bit of code knowledge. But Face Fusion is pretty powerful. And the real nerds, the real geeks out there will use something like Stable Diffusion with Dream Booth, and they'll train it on 20 images of themselves. And actually, that's probably the best way of doing an accurate face swap. But it takes a lot of time, a lot of knowledge, and like I say, right now I'm just using the easiest version and it works incredibly well.

Just so people understand. How good does it look when it swaps out the face? I mean, is it swapping out your hair too, or is it just your face? I mean, help people understand what's getting swapped here.

Yeah, it generally does everything from the hair right down to where your neck start begins.

Okay.

It will even take it. So Mid Journey will generate someone that looks like me with a beard that's kind of like. Like mine. But the face swap, again, will swap the facial hair out to look exactly like my facial hair.

Does it provide multiple examples or is it just you one? And it generally looks pretty spot on. I mean, have you found it to be like, you can't really tell. Does it look like it's legitimately real, like it's you?

Most of the time I'd say 90% of the time it's legit. As with any AI tool, there are faults. You know, it's the same with ChatGPT. If we prompt it in the wrong way, we. We get rubbish back. You know, if we give a rubbish prompt to an AI image generator, it won't give us something good. There will be times when it will fail. Sometimes it's obvious because like, maybe the face was off to the side and I gave it a front on photo of me so it couldn't quite get that. Other times I think that image is absolutely fantastic. That's going to be a brilliant face swap and I'll do it. And it makes me look like a ghost or something like that. And it's like, I can't explain that. That's just AI glitchiness, isn't it? So you have to try again or get a different image and try again.

Well, Mike, first of all, thank you for sharing all of your insights that you've learned on this journ journey on utilizing AI to create really cool thumbnail ideas all the way to the actual creation process. If people want to see with their own eyes the work that you've done, how do they get to your YouTube channel? And if they want to connect with you any other way, where do you want to send them?

Oh, thanks. It's been a real honor and pleasure to be on the show. So if you want to find me on YouTube, you can type in the name of my channel, creator magic into YouTube. That's creator as in content creator. So Creator Magic. And if you want to find me again, I have another channel that's a bit bigger than Creator Magic called Mike Russell on YouTube. So it's my name and you can pretty much find me with that name. Actually, the handle I use in most places is Imike Russell. So Mike Russell, all one word with the letter I in front of it and you'll find me around.

And if people want to check out Music Radio Creative, where do you want to send them?

Musicradiocreative.com thank you so much, Mike, for.

Coming on the show and sharing your insight insights with us.

It's been a real honor. Thank you for having me. And yeah, I look forward to seeing where this takes us. AI is certainly very exciting.

Hey, if you missed anything, we took all the notes for you over@socialmediaexaminer.com a10 and also, would you let your friends know about this show? You can tag me if you want. I'm Elzner on Facebook and ikestelsner on X Twitter and be sure to follow this podcast on your favorite apps so you get all the future episodes automatically downloaded. And be sure to check out our other shows, the Social Media Marketing Podcast, the Marketing Agency show, and the Social Media Marketing Talk Show. This brings us to the end of another episode of the AI Explored Podcast. I am your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. Best the AI Explored podcast is a production of Social Media Examiner, AIbusiness World early bird pricing ends on January the 16th. You can save $300 on your tickets. Register at AIbusinessWorld live today.