Why Your AI Images Look Like Everyone Else's (and How to Fix It)

Hey, it's Mike Stelzner. What if you had an AI system that wrote your ad copy, generated the images and video, reviewed its own work and made revisions all customized to your brand without you ever touching it? That's exactly what Caleb Cruz is teaching AI Business Society members to build on July 16th. No coding required. Join today at socialmediaexaminer.com/ai. And if you're listening to this in the future and you missed it, the full recording and resource kit are waiting for you. Visit 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.

The AI Overlord 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 put AI to work. Today we got a doozy of an episode and we're going to transition over to it right now. Today we'll explore creating quality AI images. My special guest is an AI image expert and former Ulta Beauty creative director who helps marketers and entrepreneurs with creative design strategy and AI. She's known online as Your AI Auntie. Her course is Foundations in Fauxtography, spelled F-A-U-X. Her membership is Auntie Up. Lauren Devine, welcome back to the show. How you doing today?

I'm doing good, Michael. Thanks for having me.

You know, we got a chance to talk a lot at Social Media Marketing World, and obviously there's a lot that's happened since you were last on the podcast, especially in the world of imagery and AI and stuff. So what do you see right now as one of the or some of the biggest false beliefs or misconceptions that people have about AI images?

Yeah, I think one of the biggest ones right now is that everything that AI makes is AI slop, if you will. And I think that we are sort of judging AI imagery on people that don't really have too much knowledge around what a good image or good design looks like. And so it's almost like we're judging, you know, a 5-year-old that's playing piano and being like, oh, this is the best that's available and this is the best I could ever hear. But it's like, no, Mozart exists. We just are judging it based on this kid, you know, smashing keys at a dentist's office. Like, it just isn't the same. And so we have to kind of open our brains to the idea that there is really good AI out there. It's just a matter of where are we seeing it and are we even noticing the good AI because it is so good.

Well, and what's the reality right now? Help people understand kind of the quality of the images coming out of AI these days.

I mean, it's so good. I got into the AI space. I started playing with Midjourney in late 2022, and even DALL-E, which was ChatGPT's image model, in summer of 2022. And when I was playing with DALL-E, it was so bad that I, I just basically ignored it, and I was like, this is not anything. And then I started realizing, oh no, it's going to continue to get better. And now we— here we are, 6 years— 6 years— 4 years later, and, you know, a lot of the time it is very indistinguishable from real photography. And so We used to be able to easily pick it out. We could see, oh, this hand has 6 fingers, or the uncanny valley in the eyes, or the skin looked somewhat plastic. And now a lot of those AI image tells are no longer, and it's much harder to be able to distinguish the one from the other.

Yeah, it's indistinguishable. So, okay, one of the tools that we're really gonna focus in on today is specifically from OpenAI, ChatGPT, and GPT-Image. 2.0, and we don't know if that 2 is gonna turn into a 2.5 or whatever by the time this comes out. But when you are using this leading edge image creation technology right now from ChatGPT, what is the benefit? What is the upside specifically to my audience of marketers that are out there?

So previously a lot of the image models were diffusion models, which means they kind of started with noise and then they almost like chipped away. Almost think about it like a sculptor chipping away to create this final image. And so it like denoised the image to get to that final product. And these models were trained on millions or billions of images plus their text descriptions. And so they really learned what things were. And it's not like they have some library saved of like all of the images that they've looked at. So when someone says, I wanna see, you know, a cat on a surfboard, it's not going and finding images of cats and picking, finding images of surfboards and like pushing them together. That's not how it works. It is learned statistical patterns of pixels so that when someone says a cat, it knows what a cat looks like, and so it can create something entirely new. And so these diffusion models were trained that way. But ChatGPT's new image model, and they haven't fully come out and said whether it is a diffusion model or whether it is, um, using something else, but what it— they have said is that it is backed by a large language model, the ChatGPT language model, which means it can actually think and process. So it has this thinking layer on top of it. So when you say, I wanna see a cat on a surfboard, it's able to actually think through, and you can even have it go out and pull from current events. So you could say, show me a cat on a surfboard based on what's been going on in the world the last week. And it can go and do that research and then pull back and say, okay, this happened and this happened and this happened, and now we're gonna put that into the shot. So it's actually able to process and pull from current events to help you create your images rather than just being based on its original training data. And the other piece of it is that it can actually see. So we'll get into reference images, but it can actually see and analyze a reference image, whereas these previous diffusion models, if you gave it a reference image, it almost used it as like a style and it couldn't actually see see it, but if I were to give it a image of this can of peach pineapple mango bubbly that I'm drinking right now and say, analyze this and create a world around it, it could read that it's peach, pineapple, and mango and be able to make the world in, you know, peach and pineapples. Whereas before I would've had to be very explicit about what I wanted to see, whereas now it can do a lot of that kind of creative strategy work for you.

Well, and let's also talk about text, right? Because a lot of marketers are not just using images that look like photographs, they're actually doing a little bit of layout as well. So talk to me a little bit about— because it used to be really bad with text. I mean, ChatGPT was horrible with text. Most of them were, and some of them still are. So what is it about the text layer? Because I think that adds a whole new dimension to this, doesn't it?

It does. And I think that's another piece that a lot of people are starting to kind of call the AI slop flyers that we're seeing, the ChatGPT slop flyers, because everyone realized that with this new model, it can get text really, really accurate. And you can put in a lot of text, like paragraphs worth, and you can be very descriptive about where you want that text and the type of fonts that you would want it to use and the style and the composition of it all. But most people are not doing that. They're just saying, make me a flyer, here's the information. And so because of the way that AI works, it defaults to average and it defaults to basic. So all of these flyers, while yes, they have the right information, and that's why a lot of marketers are using them now because it's a lot faster than going into Canva and trying to find a template and lay it out and all that. It's just like, oh, well, I can just tell it what needs to be on it and it'll make it. But the problem is that they're all starting to look very similar because it's defaulting to those same average fonts and it's defaulting to those same, hey, let's put a million different icons to call out every little thing. But, and we'll get into it, if you were to prompt more specifically and say, hey, I want you to use an extended sans-serif font, all caps, make sure that it's across the top third of the layout, you know, and get really deep into how you want it laid out and designed, you can get some really incredible outputs from GPT Image 2.0. Whereas with these other models, it's kind of hit or miss. Nano Banana from Google is pretty decent with text., but of all of the models that I've used, which is most of them, GPT-2.0 is going to get your text as close to what you put in as possible. I very rarely see issues with the text being wrong.

So let's talk about, before we get into the details here, let's just talk about some of the practical applications you're seeing or your clients or students are using these for, because, you know, a lot of people they just think, okay, you know, you talked about this mango universe and stuff, that sounds really creative, but from a business perspective, like what are some of the use cases now that are possible with all these images?

Yeah, so in that case where I'm talking about this mango world, that would be a great use case for a CPG brand where they've got a lot of different flavor. They have one product, but they've got 10 different flavors, right? So now they could come up with a prompt that is going to get detailed in the terms, in the way that their images come out but it reads each product's flavor and is able to make a different variation of that same shot, but specific to that flavor. Or another example, my dad and my brother have a company called Club Critters, and they're 3D printed animals, and they have like 800 different SKUs of animals, and they need to create imagery for each of these different SKUs. And so I can come up with prompt templates for them so that each animal, they can input the reference of that animal. It'll look at what the animal is, it'll say, okay, this is a pig, or, oh, okay, this is a raccoon. And it'll be able to look at the colors and then it'll be able to create a 3D generated environment. So it almost looks like the 3D printed animal is also in a 3D printed environment specific to each animal and its color and its, where that actual animal would be. And it's like, cool, I only had to write one prompt and now I'm able to use that on 800 different SKUs and come up with 800 different animals. But all of the images look like they live in the same world, right? Because we've used that same kind of template for it. So I think for a lot of companies that have product-based businesses, this is going to be so helpful because the amount of content that we need now is insane. And to be able to run ads regularly, you wanna be able to update that content, you know, every 6 weeks rather than, you know, only being able to have a quarterly shoot each year. Now it's like, okay, cool, I can get new images every week. Week if I wanted to. And that really keeps your product looking fresh. You've got new flavors or you've got new colorways that come down. You can easily just update that without having to have a full brand shoot for your, for your business anymore.

And for the B2B businesses that are listening, something I've been experimenting internally for our AI Business Society, because we're selling an experience which doesn't have a physical product, was we used Claude to kind of help come up with a reference prompt of a woman in a slight anime look working typically in a home office with that sunset where the golden hour colors, you know, which often imply like this kind of like futurist, like aspirational kind of thing. And what we've been able to do is every time I write an email, I can take these prompts directly into ChatGPT and generate the same woman in all these different shots. And now I've got visuals that kind of metaphorically go with the email, for example, we were talking about how you could mirror yourself with AI. So we had this prompt of this woman looking in a mirror and there was like an AI equivalent of her in the mirror in her home, you know, and these are the kind of things that allow you to get really, really creative for things that don't even involve physical products. And we're just scratching the surface. You could probably use these in presentations. You could use these on landing pages, all sorts of other applications, right?

Yeah. And I think that's one of the biggest things about using AI now is being able to have this cohesive brand style. And similar to what you're doing, I've got a sub-brand called Ante Up, and the whole vibe is casino, like a futuristic casino. And so I've been able to use AI to build out my entire sales page for it because everything feels like it's in this casino with robots and it's got the same colors and the branding. And so once I've kind of come up with that, I can create like a sort of style guide image to use as a reference. And then every time I'm prompting, all these images feel like they're part of this same casino world. And of course it's not real, but it lets people feel like they're part of a community or they're part of this brand and they're coming into this experience. And you can use it for your email. Like you said, you can use it for emails, you can use it on social, you can use it everywhere and everything feels the same. And so once you've got that sort of set look and style, you just use it as a reference or use it in your prompts for all of your images and everything looks and feels the same.

Okay folks, hopefully you've got a taste of what's possible here. And I and Lauren are both believers. I'm a former art director and Lauren obviously is an artist. Visuals matter. You know, in a world where AI can create great readable text, it's the visuals that are gonna really stop the scroll. It's the visuals that are gonna get people to understand what the essence is of what you're talking about. So that's why I'm excited to get Lauren here. So now, before we start using GPT Image with ChatGPT 2.0, is there any stuff that we need to start considering? Because there's a really good chance that the people that are listening right now are not doing the things that you're doing, Lauren. So let's just start with like, what are the things that we ultimately need to consider before we get started?

First of all, what we just kind of talked about is knowing what your brand is meant to look like or what it's meant to make people feel, right? Having that kind of north star is gonna be really helpful, but also being able to, articulate that, right? So you can think it and feel it, but if you're not able to explain it to an AI, whether that's the large language model or the image model, it's not going to be that helpful, right? So understanding that taste and articulating that taste are two different things. You can say, oh, I know what looks good, but if you can't explain what it is or why you want it to look that way, that's the missing piece that I think a lot of people—

how do we do that if we don't know how to do that? Because some people have no clue how to articulate their taste. Do you have any tips on that?

Building Taste is a muscle. So it, you know, you have to look at a lot of work and you have to be able to say, okay, like, do I like this or do I not like this and why? So I do think it does take some work. You have to spend time and sit with imagery and sit with design and look at it. But I've actually created a custom GPT/Cloud skill where you can give it images and you can have it break down those images for you and help you kind of understand why something might work and why it doesn't. Because a lot of people, if you haven't, you know, I went to art school, I've been designing for 30 years, and I was on, you know, photo shoots and video shoots. And so I was very immersed in that world, but a lot of people are not. And so they don't necessarily have that same sort of eye to be able to make those decisions. And so using a tool that is able to kind of mimic an art director or someone that does have that sort of knowledge is a great way to sort of use it as like an accelerator sort of in your back pocket where it's like, okay, here's a bunch of images. Can you help me describe like why I probably like these. What is it about them that I'm attracted to? Is it the light? Is it the angle that they're shot at? Is it the perspective? Is it the color? And it'll try and pull out the different parts of that image for you to help, uh, you kind of understand maybe why you are, you know, drawn to it. Yeah.

And as a former art director myself, back when I used to have a creative agency, when you're working with humans, and it's the same, like artificial intelligence is just like an artificial version of a human. You can start with some sort of initial concepts that you think are really beautiful, right? Like inspiration boards almost, right? Like you see these on HDTV and stuff. And you can say, "I really love this and I really love this and I really love this and I want it to be a little bit like this." And then once you see something you like, I would imagine you could ask AI to describe this in a way that is a repeatable way of doing it, right? Because we can do that right now. I know with Gemini, we can drop a bunch of images in there and we could probably ask it to reverse engineer and what my style is. Is that fair to say?

Yeah, absolutely. And that's what I've trained, like, this skill to do. But I've gone in ahead of time and kind of told it, hey, these are the roles that you should be playing. I want you to think like a creative director. I want you to think like a director of photography. I want you to think like a light— like a lighting expert and a stylist. Because when you give it those roles to play, it's able to think about it from that perspective. And so that's where you're going to get even even better outputs than if you were to just use, you know, regular ChatGPT and ask it. So giving it those roles and then also giving it examples of what you think is good, because it needs that context. And so that's why you were talking about like, hey, go and pull the swipe. If you've got a Pinterest board that you've created, screenshot it or grab a bunch of those images and give those to ChatGPT and be like, this is what I like, this is what I'm, you know, drawn to, and these are the sorts of style that I want to incorporate into my prompts. Can you help me now create prompts that have this same sort of aesthetic vibe and feeling.

Love it. And we'll tell everybody where you can get Lauren's skills and custom GPTs at the end of the interview. Okay, so let's just assume we've agree that, hey, if we can figure out some sort of a design taste concept, it's gonna be a moat, right? Because it's gonna be unique, something that's just our brand and just for our product. And some people already have that, right? Because maybe their brand's been around for a long time, like Social Media Examiner. So once you've got that, like where do we now technically begin? Let's say we're ready to start creating some actual images inside of ChatGPT. What, what's the basic stuff we've got to start with?

So we can even use Social Media Examiner as sort of the example. You could take a bunch of, like what we just kind of talked about, you could take a bunch of your imagery that you've already got and drop it into chat and say, hey, I want you to analyze this to help me create new images using this style. So from there, chat can write you new prompts and you can go in and say, hey, I need an image that is for social media of this woman, you know, holding a camera, taking a picture. And so if you give it a little bit of that plus your references, it's going to be able to then write you an im— a prompt that you can drop in. But what would make it even better is being able to take that prompt plus the actual image as a reference so that it has the actual colors and style that it can kind of see. Because we talked about how ChatGPT GPT can see these image, or GPT Image can see these images as reference. So when you pair a good prompt with a reference image, that's gonna make it even more powerful. And so you can drop both of those in the chat and say, okay, now make me an image, and it will be able to use both the prompt and the reference to make a new image for you.

What type of images are best as reference images? Because obviously there's a lot of images is probably that people are thinking about and they don't know what kind would be the best for reference.

So it sort of depends on whatever it is that you're creating. So if you're creating an image where you wanna be in it, so say I need an image of myself, I would wanna make sure that I've got a good image of my face. And you know, if my whole body is in it, I would want an image of my full body so that, you know, it is not making me 100 pounds when I'm not 100 pounds. And so I would want it to know what I actually look like, but you don't need like 20 different images of yourself. You need one good face image and one good body image if the body's in the shot. Of course, you know, if you're gonna have separate angle, you have to think about the fact that you're only really creating one image at a time unless you're asking for like a contact sheet. And a lot of the time that's sort of a smart way to go about it is saying, hey, I want you to create sort of like a character contact sheet for this character. So you you have one image that's got, you know, a front-facing, a side-facing, a full body, and then you can always feed that one single image in as a reference so that it always is able to pull from, you know, all those different things depending on what it is that you need. But say you're trying to create like a background for a slide deck. If you've got graphics that you use, you can drop those in as your reference image. But what you wanna be careful of is not necessary, you should not be just pulling images that you don't have the rights to and putting those in as a reference image for ChatGPT to use. So actually, just recently there's been something going around where a creator had— somebody had basically taken her exact image and gave it to an image model. I'm not sure which one, and literally just said, put my head on this body and then used the exact same image, changed nothing else about it. Same body, same background, same color, same matcha, same everything, but just with a new bot. Body.

Oh, and she was not happy. I remember.

No. And it's like, that's not your image to use. What you could do is if you like that image, give it to ChatGPT and say, hey, I really like the style of this image. Could you help me reverse engineer this so that I've got a new prompt and then change those details? Rather than having it be a white tiled background, swap it for your brand colors. Rather than having it be like, you know, Adidas Breakaway pants, switch it to, you you know, barrel jeans. Like, make it feel like it's your own because there's no point in just ripping off someone's image. Someone's gonna see it and they're gonna be like, that's mine. Like, that is my photograph. Because AI is so good at recreating an image almost to a T. And that's why ChatGPT Image is really good at being able to edit images. So if I had a real image of myself, I could drop that in as a reference because it's my image and say, hey, change the clothes, or change my hairstyle, or change the background so I'm in a different seat. But those are my images to use. You cannot just take someone else's images that you don't have rights to and try and, you know, drop your own product into them. That's not the vibe.

You know what most marketers are doing manually? Ad creative, writing the copy, coordinating with the designer for the images, waiting on video edits, managing revisions, revisions, every campaign, every cycle. I wanted to fix that for our members. So on July 16th, I'm bringing Caleb Cruz to teach inside the AI Business Society. He's going to show members how to build personalized AI pipelines using Claude code that runs the entire creative process on its own— copy, images, video, revisions— all customized to your brand. And you don't need to be a coder to to do it. I'll be in the room for the entire session. I'm in every session, and I can tell you members are going to walk away with a clear, buildable system that they can put to work that very same week. If you can't make it live, the full recording and resource kit will be waiting for you— transcripts, slides, key takeaways, the whole shebang. This is what we do every month inside the AI Business Society. Come and see for yourself and don't miss this training. Join at socialmediaexaminer.com/ai. Again, socialmediaexaminer.com/ai. Going back to the kinds of images we talked about, if you're taking a picture of yourself, you know, you can do that. What about things like product logos and product shots? Should some of those things also, especially if we have products, you know, and logos, should those be references as well? Just what are some of the basic references that we might want to consider because otherwise AI is just gonna probably make it up, right?

Exactly. Exactly. So if you have a brand and you wanted to use your actual logo, you need to give it that, um, because it will go and make that up. However, if you are a more well-known brand, say you're Adidas or you're Nike or you're Olipop, it's looked at so many pictures and it is able to like go out to the internet it knows what the Nike logo is, it knows what the Adidas logo is. So it's got those. But if you're a smaller, newer brand and it doesn't have, you know, tons of data on your logo, it needs that to understand what it looks like. And you need to be specific in your prompt and say, hey, don't change anything about this logo. This is the exact logo that I want you to use. Keep it exactly as is. And that's going to make sure that it, you know, sticks to what your logo is. Same with like brand colors. A lot of the time you can feed it like your actual hex text colors so that it knows to use these very specific brand colors rather than saying, you know, blue and orange. It's like, okay, but what blue and orange, right? So you can get very specific with it now. Previously it was much harder to get it to do that. Now the models are very— I mean, specifically GPT-2.0 can absolutely take reference images. If you've got brand patterns that you use, you could drop those in. But the thing that I would say is don't just give it stuff to give it stuff. Give it the stuff that it needs for that specific task or that specific output that you're looking for. But if it doesn't need that, don't give it to that. You know what I mean? If you're asking for an image of just like a full, you know, a picture of someone's face up close up, it doesn't need a full body image. So don't give it that because then it's almost gonna be like, well, why did you give me this? Should I make an image that is full body or should I make this close up? So make sure that you're giving the specific things that it would actually you actually need for the image you're making.

So we're about to get into how to prompt it, but real quick for the Illustrator friends who are listening, does it work with vector, you know, EPS, Adobe Illustrator, SVG files, or do we just need to give it like flat, you know, PNGs and JPEGs and stuff like that?

Usually it's just JPEGs or PNGs. I work in a different tool and we'll get into that. I don't necessarily work in ChatGPT ChatGPT as much when I'm creating images. You can, of course, but from my understanding, I don't know that it would be— it probably would be able to read an SVG, to be honest with you. ChatGPT would be able to, but it's honestly simple enough that you can screenshot stuff. Like, as long as it can see it, that's what it needs.

Love it. Okay, so we've got a bunch of references now. We're going to be using those somehow, some way. So like, where do we actually start now that we've got these references that we're gonna be, uh, including.

So you would literally just upload those references and say, hey, I've uploaded these reference images, here's the prompt, and run it. Or you can say, hey, I've added reference images. This is the type of image that I'm trying to create. Could you please act as a, you know, photographer and help me write a prompt for an editorial image of, you know, this product in a flat lay environment and you give it a little bit and then let it kind of help fill in the blanks. 'Cause we've talked about with the fact that it can think and analyze, it can kind of potentially come up with a better prompt than maybe what you would've written on your own. So giving it a little bit now is sometimes enough, but the more detail you give it in your prompts, the better it's going to be. But yeah, that's as simple as it is. Give it the references, give it the prompt, give it what you're trying to create, and then tell tell it, you know, generate image, and it will just generate that image right there in your chat.

So when you were last on the show, you went really deep into what I'm about to talk about next, but why don't you do a flyover of kind of the key things you want to include inside the prompt to make sure that the AI kind of knows what to do?

For sure. So this is my 7-pillar prompt framework, and I'm gonna go through all 7 of them, but don't think about them as like a checklist of you have to have every single one all the time. Think about it more as like a control panel. So the more detail in each of the different pillars, the more it's going to follow it. If you like, these models are so good that any detail you give it, it's going to do that. But any detail you don't give it, it has to make up. So you think you might have a really long detailed prompt, but you're still leaving out a million different, as a different, you know, pinpoints of what that image could be. So the first thing is the medium. And that's what's the type of image that you wanna get back? Is it a photograph? Is it an illustration? If it's an illustration, what type of illustration is it? Like a Sharpie illustration? Is it a fine art illustration? Is it a kid drawing with crayons? So getting specific around that. Do you want a logo? Do you want a 3D render? So any sort of, you know, what is that thing that you actually want? That's your medium. And then the next is gonna be your subject and your action. So what is the What is the actual thing? Is it like we talked about a cat on a surfboard, or is it a can of soda, or is it a person looking in the mirror? Um, but what are those things doing? So rather than just a person, it's a person looking into a mirror. Maybe how do they feel when they're looking into the mirror? Are they looking contently? Are they looking angrily? Are they punching the mirror? Are they standing straight? Right? So what is it? Is the can balancing? Is it dripping wet? Is it— what is the thing that it's doing? Because that's what's going to help it tell the story. Um, the next is the setting or the scene. So what's actually happening around the subject? You know, what's the environment? Describing that as detailed as possible. You could say, you know, a retro diner, or you could get really detailed and say a retro diner with wooden panels and neon signs for Miller Lite and like As much detail as you put in, it's going to give back. Anything else, it's going to default to whatever the average retro diner looks like. So you're gonna see, you know, a jukebox in the corner and you're gonna see, you know, brown, like it's gonna just default to that. So give it that specification. The next thing is going to be your composition. So how is the image or the graphic composed? Is it a wide shot? Is it a closeup shot? Is it shot from a Is it shot from below? Um, you know, if it's a design, is it symmetrical? Is there a lot of text? Is it minimalist? Is it maximalist? Are there 3 columns? Like the more detail you give about the way that the image is composed, the more original it's going to be. Because if you don't give it anything, it's gonna be a straight on shot, right? It's just gonna be like, what is the most average basic shot that I could give you? The next one is going to be the lighting. So this is gonna be more specific to photography sort of things, but You know, is it warm lighting? Is it cool lighting? Is it light coming in from, uh, a curtain at daybreak? Is it, um, a small light in the corner? Light will really change the mood of an image. And so the more detailed you can get about the lighting, uh, the more, again, you are going to be creating a more original image for yourself. Um, the 6th thing is going to be the aesthetic or the vibe. So this is where you can kind of bring in Ralph Lauren references to different eras, different art styles, different designers or artists. Now, a lot of people will say, hey, don't put artists' names in there, but you can absolutely, you know, say like, hey, I like the vibe of Wes Anderson. Can you help me understand why, you know, the style and the aesthetic? And then pull that in. Very symmetrical, very color poppy, right? And so you're going to be able to kind of pull in the vibes of what he does. It doesn't mean you're trying to rip off Wes Anderson. It just means you like that style. And that's the way that most creatives work already is they go and pull, you know, examples of screenshots from his films and say, this is sort of the direction that we're going in. This is the same thing. You're just kind of describing it. And then the last one is going to be the intent. So what is the intention of this image? Do you want people to feel, you know, if it's a skincare ad, do you want people to feel the efficacy? And or do you want people to feel, you know, that they need to buy this immediately and it has to happen fast? Or do you want them to feel like they need to be a part of this brand? Like any sort of little direction you can give it around what's the point of this is going to make the image better. Because again, GPT Image 2.0 can actually analyze and think about these things rather than, you know, if we had done this 2 years ago, the intent really wouldn't have been that big of a deal. But now that it can think about it, it actually can kind of input that sort of feeling into your images.

It's fascinating to think that the intent matters because from a visual perspective, it's not clear what that really means. But because this thing is powered by a language model, it understands the emotional objective. Like, if the intent is to generate concern to get people to move off the dark, well, then that might mean the expressions on the people's faces might be different, right? Or if the intent is to, to show transformation, you know, then you might show people smiling and you might have certain kinds of light and the way the light works and all that kind of stuff, right? So that stuff really, really matters.

Okay.

You hinted earlier that there's more than one way to use GPT Image 2.0, and we've been talking about how to do it using ChatGPT, but there's another way to do it. And what I would love you to do is kind of explain your preferred other tool, and let's kind of talk about the pros and cons, just because that's important and not everybody really even knows what we're about to share.

Yeah, so I use a tool called Magnific, formerly called Freepik. Magnific was actually an AI image upscaler, and Freepik they bought them, and then Freepik had originally started as stock, like graphics and things like that. And so as they became more of this AI tool, they were like, Freepik isn't really a great name for us. Let's just use Magnific since we own them now. It just sounds better. And so they recently rebranded to Magnific, and it is a, at this point, an AI tool aggregator. So what that means is means is we've been talking about GPT Image 2.0, but there are over 40 different image models that you can access within this tool, as well as video models, audio models, 3D models. And so Magnific is sort of a— my go-to for anything that I'm working on that would require AI. And so I can easily create an image and then turn it right into video right in that same tool. Or I could go and create an image and turn that into video and then add in a lip sync in there. But the other piece of this is that if you're using it in ChatGPT, if you're just creating images in ChatGPT, you're gonna get one image back unless you turn on like extended thinking and you ask specifically for, you know, 3 or 4 different images. Or you can, I think it'll generate up to 10 images in ChatGPT, but you have to be specific about it. You have to ask for it. And a lot of the time you're not getting variations on that image. And that's what I think is really important for people to understand is that every time you prompt any image model, you are getting a new original unique image. I can give the same prompt to ChatGPT 5 different times in 5 different chats, and every single image is going to look different. It's almost like that spot the difference game that we used to play in highlights where it's like, oh, the can is a little bit different here, or oh, the background Oh, the background is a little bit different there, or, oh, this, you know, the way that the typography came out is a little bit different. And so when we're only getting one image back in ChatGPT at a time, it's not as helpful because it's not perfect. And I think that's another thing that I should have spoken about at the beginning around like what's the misconception is that it's so easy. You just push a button and you get a perfect image back. And that's not the case. A lot of the time there's something weird, something is not perfect.

Perfect.

And so, or the lighting isn't exactly the way you want it, or the product name didn't come out perfectly each time. And so you have to generate again, and then you have to generate again. And so rather than having to just keep hitting generate when you're in ChatGPT or starting new chats, when you're in Magnific, you can run up to 8 images at a time with one prompt. And so I can run that one prompt, and now I've got 8 images to decide, okay, which one is the closest to what what I'm thinking. Or the other reason that I love it is that a lot of the time, if I'm not giving it that much detail, or even when I am giving it more detail, there are things that I wouldn't have even thought of that show up in the image. And you're like, oh, I really like how it did that. And then you're like, that's what I want to do with the next roll, the next roll. And so you kind of say, okay, let's pull this as a reference image. Now we've got this new image that we created, but now I want to make these changes to it, and I can make edits to it. So having more choices as a creative, I do I think it's a double-edged sword sometimes because it's like, okay, yes, I've got options, but sometimes when you're ADHD, you're like, there's too many options. Like, take us, take some of them back because now I can't decide which I like best. But I do truly believe that when you're working with AI, you need to have these different variations to be able to look at. And, and that is the piece that is really important because taste a lot of the time is saying no, not this, right? And being able to pull out the things that work and the things that don't work and find that true winner from the options.

We're going to talk a little bit about how to actually take advantage of Magnific. But for folks that have been listening for a little while, back in June of this year, we had Jared Lew on the show and he was also recommending Magnific for AI video. So this is the second time someone has recommended it. Just real quick, explain kind of what the costs are and then let's get into some tips on how to actually create series of images because ChatGPT can be pretty darn slow sometimes creating images. So let's start with just kind of like what the kind of costs are that we're looking at for a tool like this?

Well, it depends. So there are a bunch of different tiers for Magnific. I always tell people I would not go with the lowest tier, which is like the premium plan, because it really isn't that many credits. And because, you know, I'm telling you guys, you need to generate more images at a time, I feel like you'll run through your credits or you'll be very much kind of on the, hey, let me look at what, how many I'm at, and then you won't generate more because you're like, oh, well, I'm close to, you know, hitting my limit for the month. But then the thing about Magnific is that you do have tiers where you get unlimited images, but it depends on what model you're using. So none of the, the tiers actually give you unlimited using GPT, but they do give you unlimited using the Nano Banana models. So that's the other thing is you can run images using GPT and run them using Nano Banana and be able to compare because all of models are better at different things. And so to be able to just kind of flip between them and be like, oh, I actually really like this one, you could use that as a reference then with GPT, you know what I mean? So you can kind of mix and match.

But so what are you paying a month?

I'm not the right person to ask for this because I'm, I generate so much. I think I'm paying like $2,000 for the year, which in terms of the number of images that I've created is, it's kind of nothing.

It's kind of nothing for someone who this is their job. Right?

I mean, but that's probably not what most of you guys are gonna be paying. I would say probably it's like $50 to $60 a month to be able to get like a good amount of images. And when you are on one of those unlimited plans and you're using the other models, if you're, if you're generating images in Nanobanana, you're being able to save your credits to use for GPT or use for video or use for upscaling. So it's kind of like a, two in one hand, one in the other, if that's the saying. But I definitely think it's worth it if you're serious about AI image generation to be doing it in a tool like this because you've also got all of your images in one place. Whereas, you know, if you're ChatGPT, it's like, oh, you've got some here and then you've got some in this chat. Of course you have like an image library, but it's just, it doesn't function the same way. And now the other piece about Magnific that's really, really awesome is that they have MCPs now. So an MCP connector to Claude. So now you can just, you don't even have to be in Magnific, you could just be working in Claude and it'll generate all the images right in your chat for you. So it's kind of crazy that you're using GPT's image model inside of Claude. And they've also just introduced in the last week plugins for Magnific for Adobe Photoshop and Adobe Illustrator. So now I've got a plugin in both of those tools and I can literally pull up any of the images that I've created in Magnific right in Photoshop. So I don't have to go and copy it or download it from Magnific. I just pull it right there. And then I can bring in images that I've created in Photoshop directly into Magnific to work with and use in AI. So they've really made it super, it's really, really helpful. They've got a Figma MCP. So it's being made to make it so that all creatives can be using it in all of their different tools, which makes it very, you know, agnostic.

Give us a couple tips on how to use it. I mean, do we just say create images and just use the same prompts, or is there a couple of quick tips that you could share?

Yeah, so I think what I like most about it is that because I use my Prompty Poppins skill all the time, and we can kind of quickly hit on skills, skills are basically a repeatable process. So I've gone through the process. I just talked to you guys about the exact way that I prompt, but in there I get very detailed about the specifics. I give it roles, I talk to it about which models it should use depending on what it is that I'm creating. And so I can just go in and say, hey, can you help me write 5 different prompts for, you know, this product, this coffee product that I'm creating images for? I give it it, it looks at it, it writes the prompt in the exact way, in the exact way that I want, all 5 of them. And then I say, okay, just go and generate 'em. And so then it goes and generates them right there. There rather than me having to go copy the prompts from Claude into Magnific and then generate there. I can just say, hey, I want you to use this model, make me 3 different variations in this model for each of these prompts. And it'll come back and spit back, well, I can't do the math, but whatever that number ended up being.

So wait, just so we're clear, you can create a skill and we're gonna tell 'em how they can get your skill, but you can create a skill and you can import the skill into Magnific is what I'm hearing you say. Is that correct?

You're not importing it into Magnific. You've basically connected your Magnific account to Claude. So that when you're working in Claude, you're using your Claude skill to write the prompts for you and know which models to use. And then it just goes ahead and you say, okay, create the images. And so it shows them to you right there in your Claude chat. But then when you go and open Magnific, they're there too in your library. So it's basically like you're controlling Magnific via Claude. And so you've got the creative, you know, strategist skill to do the prompt writing for you.

And just so everybody understands how this would work, you'd have to use either Claude Code or Claude Cowork to be able to pull this off?

No, no, no.

Oh, the Claude. Wait, you're telling me you can just use the good old-fashioned Claude Chat?

Yes, you can.

Because these connections and it will just bring those images right back into Claude?

Yep. Woo!

You gotta be kidding me.

Yes, yes, it's very exciting. Now, if you were to be using it in Cowork, then it would potentially be able to like save the images directly to a folder on your, so like when I was building my sales page for Ante Up, I was able to basically be like, Okay, here's what I needed to do. It was building the website for me and then I was like, I want like a top hero image of like a neon sign with like a robot. And then it says everything. It went and wrote the prompt, generated the image, and then dropped it in. And I was like, okay, this looks okay, but like I need there to be motion. And so I was like, okay, hold on. It went and wrote a prompt for me for motion to then turn it into a video and then brought it in all on its own. And so it literally did the whole process.

On the motion side of things, is that using something other than GPT Image 2.0?

Yes, but because I'm using Magnific, it's got access to all the best video models. So it used Seedry or Seedance. It can use Google Omni now. So it's got all of those different models and it just went ahead and said, okay, I know what you're trying to do because we've been talking about this whole website. I know what the vibes are. I know what the colors are. I'm gonna just go and write a prompt for you because I already know from the skill how to write a prompt and it just does it. So it's like, it's nuts. Honestly, I was like shocked. And the sales page looks so good because it's got this really cool— with sound too. So it sounds like the neon sign is turning on. Like, it's very exciting. It's an exciting time to be making websites when you don't know how to code.

Folks, when I say this, I mean this. Lauren is definitely on the leading edge of imagery and AI. AI, and the stuff that she's doing is absolutely spectacular. Lauren, tell everyone where they can see your awesome work, because I'm sure you post it somewhere out on the socials. And then tell everybody where they can get your Prompty Poppin' Skill slash CustomGPT and all the other stuff you've got going on.

Yeah, so you guys can find me. I share most of my work. I share tutorials and things like that on Instagram at @youraiante. And then for anything else, I've got youraianti.xyz. So there's going to, you know, you can go there to be able to grab the skills, but you can also grab them in my link in bio on Instagram too, if that's easier. The skills work for both Claude and ChatGPT. So if you're still a GPT user, I've got a custom GPT that does the same sort of thing for you. But yeah, that's kind of a where am I. I'm always on Instagram. I can't get enough of Instagram, but I'm also gonna be running a challenge that starts in September, end of August, that is going to help people learn how to use Claude to help them come up with a brand strategy in the first half of the challenge. And then the second half of the challenge is going to be how to come up with all of your brand visuals using Magnific. So you're learning in 21 days really how to use AI to do brand strategy and brand visuals. And you walk away with all of that, your brand, your full brand strategy, your full visual world, and and understanding of how to use both of these tools, which is pretty cool.

Where do they find out more about that?

Challenge.yourauntie.xyz.

Lauren Devane, thank you for— I could spend another hour with you. She's amazing. Lauren, thank you so much for coming on the show and sharing your insights and wisdom with us.

Thank you for having me back. I appreciate it.

Hey, if you missed anything, we took all the notes for you over at socialmediaexaminer.com/challenge. A118. Be sure to follow this show on your favorite podcasting app. And if you've been a listener for a little while, we would love a review on whatever platform, but especially on the Apple Podcasts side of things. And would you let your friends know about this show? You can tag me on Facebook, LinkedIn, Instagram, and X. I am Mike_Stelzner on X and @Stelzner on all the other ones. And do check out, uh, my other show. The Social Media Marketing Podcast. This brings us to the end of the AI Explored Podcast. I'm your host, Michael Stelzer, and 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.

The AI Explored Podcast is a production of Social Media Examiner.

Before you go, if you wanna stop grinding through the ad creative process manually, manually, the AI Business Society can help. Caleb Cruz is teaching members how to automate the entire process on July 16th. Full recordings and all the resources are included with your membership. Join at socialmediaexaminer.com/ai. Again, socialmediaexaminer.com/ai.