NanoBanana vs ChatGPT vs Midjourney for Small Business

Transcript

Kai: Can you introduce yourself?

ChatGPT: Hi, I'm ChatGPT.

Kai: Great to have you on the show. So today we're talking a lot about AI image generation. If you could give one tip to a small business owner about creating images with AI, what would that tip be?

ChatGPT: Start small and specific. Use it for rough drafts and concepts, not finished brand assets. Treat it like a sketchpad, not a printing press.

Kai: That's great advice. Anything that a small business owner should definitely avoid doing?

ChatGPT: Yeah. Don't use AI images you can't trace or have rights to for things like logos or anything legal or regulated. And don't skip human review. If it touches your brand, customers, or compliance, you want a human sanity check.

Kai: That's really good advice. Well, thanks very much for your help.

ChatGPT: Happy to help, and good luck with the episode.

Narrator: This is AI with Kai: practical AI for real business leaders. We break down what matters, talk with people using AI in real businesses, and share one idea you can use right away

Kai: Hey everybody, it's another great day to talk about AI. I'm your host Kai Gray, along with producer extraordinaire Tim Scott.

Tim: Hey, hey. Happy to be riding along.

Kai: Excellent. Well, if you're new around here, the show is for small business owners, no degree required. We just talk about AI, we share what's working, and we give you the stuff you can actually use in your business this week. Real quick, before we get into it, we've got a free guide over at aiwithkai.com. The link's in the description, and if you hit follow wherever you're listening, that helps us out a lot. All right, let's get into this week's episode. Tim, have you ever used AI to create an image?

Tim: Oh, absolutely.

Kai: What kind of images? If they're a little spicy- Well- ... you don't

Tim: have to say. Just last week's episode, I took a, a pic of you and, uh, Wes Coody on, on my phone.

Kai: Oh, that's right.

Tim: Yeah. And then I- I saw that ... dropped that image in, uh, ChatGPT.

Kai: I remember looking at that thinking- Yeah ... "Huh, that looks different. I don't remember that."

Tim: Yeah.

Kai: Very cool. So today we are going to be talking about using AI for image generation, and this is something that I think a lot of people do. Uh, but we want to get into how to do it more efficiently and especially for your business. So most business owners think AI image tools require design skill or quote unquote prompt engineering. They don't, and I want to go through that today with everyone and see how you can use it in your business. Let's think about it this way. This is not about replacing your designer for the big stuff, and this is a very important distinction. Designers are still a key component to your marketing program, but AI is there to help you with the high volume small stuff, right? Think about like the seasonal promos and the menu inserts and the we're hiring graphics, or in our case, the social media posts for AI with Kai or the LA AI social media posts, things like that. These are things that you can do very quickly that used to cost you a lot of money to do and took a long time to do it. So what used to be a week turnaround on something is now m- a minute or two, right? So you have the chance to, to experiment with a lot of different things. Quick numbers before we dive in. Nearly nine in ten organizations, or eighty-eight percent according to McKinsey's latest survey, are already using AI somewhere in their workflow. And about ninety-four percent of marketers say AI is in their content plans this year. So if you've been thinking this stuff is just for the big guys, the data says your competitors down the street disagree. And the tools keep getting cheaper, and we're gonna discuss these today. Recently, Google dropped NanoBanana 2 Lite. That's all the name of one product, NanoBanana 2 Lite, an image generation tool that takes about four seconds to generate something and is rolling out for free inside the Gemini app. Again, not a trial, completely free. We're also gonna talk about Midjourney, which just rolled out its version eight point one model, and we're gonna talk about the image generation capabilities of ChatGPT, which used to be known as Sora. Sora was a dedicated app that got shut down, and the technology w- was moved inside of the ChatGPT app. So we're gonna talk about those three things, and we're gonna get into how to create really effective prompts to get the content that you want out.

Tim: How the heck did they come up with the name NanoBanana? NanoBanana.

Kai: It just, it rolls off the tongue. I have, I have no idea, actually. It's a really good question. Right now, there's some crazy names for things, and- ... top of the list is probably NanoBanana.

Tim: Yeah.

Kai: I like when I hear serious people, you know, in businesses- ... have to, have to refer to it and, you know, it'll be serious, serious, serious, NanoBanana- serious, serious, serious.

Tim: Yeah.

Kai: It's one of my favorite things.

Tim: That's funny.

Kai: Um, all right, so we're gonna talk about NanoBanana first because right now that is the top tier. That, that's the, the go-to for a lot of people. Google came on really strong in early two thousand six, late twenty twenty-five with this, and they've baked it inside of the Gemini app. And so if you go to gemini.google.com, you're gonna, um, y- you're gonna get their, their chat window, and if you click the little plus sign, it, it opens up a dropdown, and you can say Create image. That's how you access NanoBanana. It is right now far and away the best overall tool that, um, you can use. There's three tiers to NanoBanana. There is a Pro tier. That's their flagship. Uh, it produces four K videos. It's best in class. It has the best legible text. Um, it can even create sort of paragraphs within an image and can be translated into many other languages. There is NanoBanana two, which we just talked about a little bit, and there's n- now NanoBanana 2 Lite. Those two second versions are, are-- really have to do with speed, um, and, and quality I would suggest that they're all very good. Uh, Pro will take longer, and I, I do tend to set my stuff on Pro for default, but I don't see a lot of difference between the three. So I think some of it is gonna be user taste. So when you go there and you try to create a- an image, maybe try it in the, in, you know, the Pro, Ba- Banana 2, Banana 2 Lite, and see which one works best for you. The good news is if your business is a Google Workspace account, you get this for free, you get the Pro stuff for free. Even if you're not a Workspace account and you just have a Gmail address, um, and, and you wanna use the personal stuff, you can still use Nano Banana for free. A couple things that are important. There are some limits to Nano Banana, particularly on the free side, so 20 images a day. They all have a visible sparkle watermark. If you've created an image with Nano Banana, you're gonna see in the bottom right a little watermark. The Pro version is $19.99 a month. You get 100 images a day, no more watermark. It's m- more designed for professional users. Next up, I wanna talk a little bit about ChatGPT or OpenAI's GPT Image 2, and, you know, we'll call this a eulogy for Sora. So- ... a long time ago, back a whole year ago, Sora was the king. Sora was able to do, uh, images and videos, and they were p- they were spectacular because, you know, in the dark ages of AI image generation, we had the six-fingered person. If, if you remember, like, the easiest way to tell an AI-generated image was to look at the fingers 'cause it just had real trouble with fingers and toes and things like that. Everyone would have, you know, six or more fingers. Sora figured that out, but what Sora actually did, which was the most helpful for small businesses, was that it got text correct, and it instantly became the, the, the top dog in this. And the thing that it did really well was it handled text really well. So now you could create a very photorealistic image with signage that said what you want it to, and if you had a logo, you could put your logo on things and, and it would understand what to do. And Sora, for a long time, was the, the king. If, if you remember, Sora released its own app, and they wanted to become a social media app, and everyone was creating these, like, Sora videos That, that had a, a week or two of, uh, popularity, and then it kind of died, and with it Sora died, which was really sad. Um, but at the same time, Nano Banana had, had now taken over the crown for image generation tools, and so people kind of forgot about Sora. Recently Back in April, OpenAI shut down Sora, shut down the app, and they moved all of the, the model tools into ChatGPT, the app, and it's now called GPT Image 2, but it's not something you access independently. It's something that you, you just ask Chat-ChatGPT to create an image, and it will call on the old Sora model and, and do it, and it does a very, very good job. The next tool that I wanna touch on is Midjourney. Midjourney just came out with their Model 8.1. If you used to use Midjourney and then you stopped, this is a reason to go back and try it again. If you've never, ever used it, um, I would encourage you to because I think these are the kind of things that are important to experiment with and see which ones you like best. V8 fixed Mid-Midjourney's historic weakness, which was mostly text, um, which didn't render properly. They too had an issue with fingers and toes and ears and noses and all that good stuff. But I would say that Midjourney is flat out creates the most beautiful images, right? If, if you're looking for... You think, think of, you know, desktop wallpapers or things like that- Like

Tim: the vibrant colors

Kai: nothing beats it right now.

Tim: My memory of Midjourney, uh, was early on, and I remember it was quite a process to do an image. Yes. Like, did they change it? They did. Like, is it easy now to-

Kai: They did. It, it used to be that it was only accessible through Discord- That's right ... and I mean, there was a hurdle. There was a, there was an onboarding hurdle. Um, and they've now, they've now moved away from that. Now it works like everything else, so it, it, it's much easier. Um, there, there's more kind of fine print. So for instance, if your company does over a million dollars in revenue, their pro plans do not give you commercial rights. There's, there's some other things that you might wanna look at, but for, you know, for all, all intents and purposes, um, y- you can use it, and you should definitely try it out. It's, it's really good. So a couple things to understand. Each tool speaks a slightly different dialect, right? So same brief, different prompt. ChatGPT and Nano Banana want conversational prose, as in brief it like a human artist, so full sentences. Midjourney wants structured stylistic prompts with parameters. Think of a GPT-style prompt. So a GPT-style prompt literally falls in Midjourney and vice versa, which is exactly why you let the LLM translate your brief into the target tool's dialect. So what, what this means is that- The way that Gemini or ChatGPT want things is a flowing sentence, um, think of a, a paragraph, whereas Midjourney wants things as a bullet, a bullet list. Um, so when you have AI write your prompt, it already knows that and, and can build your prompt i- in the w- in the way that it's best consumed by the image generation tool.

Tim: Wow. So each, each LLM benefits from how your prompt is structured.

Kai: That's correct.

Tim: Wow. I learned something new today.

Kai: And, yeah. Well,

Tim: good. I've just been throwing the same prompt in everything.

Kai: Here's an example of one. Um, subject plus action plus setting plus compositions plus style. So things like lighting, golden hour, soft window light, and then camera detail. Show, you know, shallow depth of field, and then color mood. I, I would say that lighting is probably the single most important thing that gets left out for all of these. It has the biggest impact if you say you want something, you know, golden hour lighting, it's gonna be dramatically better than if you, if you leave out that detail. Um, if you're a photographer and you understand lenses and, and focal points and things like that, you can use all of the same nomenclature for photography in image generation. So if you, if you want, um, something that mimics a, a 35 millimeter lens or a 50 millimeter lens, and you want certain styles that you're used to in photography, you can use the same terminologies that you use in photography. They all translate very well into image generation things.

Tim: You know, the using AI to write your prompt- Mm-hmm we talked about that in an earlier episode. Yep. Um, I think it was a playbook, right?

Kai: Yep.

Tim: I remembered that the other day. I was trying to create an image for a real estate agent. I dabble in real estate. It's kind of a family business. Shout out to my mom, Jan Scott, if you need to buy a house or a property. So she listed a interesting double lot in, uh, Old Town Daphne near the bay, and so the lots are really weird, right? Mm-hmm. They're just houses everywhere, the lines and stuff. So, so I got the plat and put it in ChatGPT, and I wanted to try to get a, uh, kind of a rendered idea of what a house would look like on there.

Kai: That's a great idea. It was amazing.

Tim: Yeah. It took a little work. Like, I was just fighting with it, right? Because it would be, like, 95% perfect. Like, the house is beautiful. I'm just like, oh, my God, like, I wanna buy this lot, you know? Right. And then the, the detached garage would be on the lot next door. And I'm like, okay. So I'm fighting with this for, like, 30 minutes. In my mind, I'm thinking, eventually it's gonna spit out something that I can use. And, and it's like, then it would get the garage right, then the house would be wrong. Right. I mean, it's just like on and on and on. And then I remembered, I was like, you know, that episode that Kai talked about, let AI write your prompt So I stopped, I went back, and I told AI, "Write my prompt for me," basically.

Kai: Mm-hmm.

Tim: And it had all of the, the previous text where I was trying to get all this information. So it wrote a prompt- Right ... based on all that 30 minutes of history of trying to get this thing right, and a lot of the verbiage is stuff that you were just talking about, you know? And so- Mm-hmm ... the reason I'm telling this story is if you're a small business owner and you wanna work on this stuff, don't be intimidated by all this prompt verbiage that you may not remember if you're driving in the car.

Kai: Right.

Tim: You can just have AI write it.

Kai: That's right. In, in fact, AI will do all, all of the things that we're talking about, sort of the structure of these prompts, in most cases AI will just do those things. You don't have to tell it to, right? So if you go to Claude and say, "Hey, I, I'm trying to create a, a thumbnail for the new episode of AI With Kai," we're not gonna have to specify the lighting or the camera angle or things like that. But if we want more control, we, we can do that, right? And, and we can tell it to create a prompt and also include these things, and it'll put it in the right area. So for instance, um, when we're talking about text, it's a very good habit to get into put, to put the text that you want in your image, put it in quotes, right? So, so you identify it to the AI as, "This is a quote. I don't want you to change anything about this," right? It's not, "Don't paraphrase this. This is the name of my company. Don't come up with a new name of the company. Um, so you put things that you want to show up, you put text that you want to show up in your image in, in quotes. Uh, a- another trick, put things in the order of importance that you want them. If you want the, the formatting to be it has to be sort of a certain size for LinkedIn or Facebook or something like that, that needs to go first, right? If that's a non-negotiable, it goes first. If golden hour lighting is a non-negotiable and it has to be in the image, that goes first, right? That, that's the idea. And, and so you may have six or seven parameters or options, put those in, um, in, in their importance, in order of importance.

Tim: Oh, cool.

Kai: I'll give you some examples, and then we're gonna, then we're gonna work on our own. So a product shot, for instance. So if you're, if you're marketing a, a product of some kind, um, a, a prompt may look like, quote, "A photorealistic photo of a mason jar of homemade hot sauce on a rustic pine table. Sliced peppers scattered nearby. Soft window light from the left. 50 millimeter lens, shallow depth of field, warm natural color grade. So- That's a lot of the things that we just talked about. Another one, let's see. A weathered cypress dock at golden hour, a streaming tray of crawfish in the foreground, Spanish moss in the trees, warm, humid evening light, shot like a lifestyle food magazine photo, shallow depth of field with, quote, "Friday Boil 5:00 PM," unquote, in bold white sans serif in the upper right So we can all, we all know what this, this is gonna look like, right? Being, you know, sort of a, um, a crawfish boil. But, um, giving it all of that direction is going to really optimize your chances of getting something that you really like on the first shot. I'll tell you two other ones that I use a lot. One, for headshots, if you need to create he- headshots, I would always recommend hiring a real photographer for really good headshots. But if you're trying to create a few different ones, uh, here's a good prompt. Quote, "Using my attached photo," this is where you'd upload a photo of yourself. "Using my attached photo, keep my face, hair, and expression exactly the same. Replace the background with a softly blurred downtown storefront. Fix the harsh fluorescent lighting to soft natural daylight, and keep it looking like a real photo, not retouched. So in that case, you, you want something that is identical to you. It's your face, and it's not adding a smile. It's not adding, you know, a quizzical look to you, something like that. It's just really changing the background. Where I use something very similar to, to that is for, uh, video conferencing backgrounds. So I have four or five where I, I will go in, I will upload an existing background that I use and say, "Oh, could you add, add this or this or this, or change the camera angle, or do something like that? And so if you're on a lot of video calls, have AI create backgrounds for you with your company logo, with the setting that you want, w- whatever you're interested in.

Tim: No, that's a great idea.

Kai: I wanna touch on very quickly commercial rights, 'cause a lot of people have questions about this, and I think this is important. You own what you make on ChatGPT, Gemini, the paid Midjourney, Canva paid. Um, basically all of the-- anything you pay for in a monthly subscription, you own what you make. So very important. However, you may not be able to copyright it. The Supreme Court let stand, uh, a case that purely AI-generated images have no copyright protection, so a competitor could reuse your ad image. So just something to consider when you're making this stuff, when you're putting it out, that you own it, but it could be reused potentially 'cause you can't copyright it. One little trick here is that real human editing and arrangement restores protection. So let's say you create an image in ChatGPT, and then you take that image and you put it into Photoshop and you tweak it a little bit. You, you now restore your protection

Tim: Okay

Kai: So just something to keep in, keep in mind. One other quick thing on water- watermarks. We talked a little bit about this for NanoBanana, but Google embeds invis- invisible SynthID, which is a tag that it can put inside of an image. You can't see it. Um, SynthID is becoming sort of the standard for this sort of thing, so they put a SynthID in everything. ChatGPT images carry C2PA credentials and SynthID. So again, two different forms of identification, but for everyone listening, just can-- just remember that if you create it in AI, it, it, it will be detectable as AI by something. Maybe not to a human, but if someone wants to figure out if something was really created by AI, it's very doable.

Narrator: Mm-hmm. The Playbook: One practical way to use AI right away.

Kai: So we're gonna create a thumbnail. So let's, let's write a prompt. So the first thing that we're gonna do is we're gonna take photos-

Tim: I was gonna... Yeah, take photo?

Kai: Yeah.

Tim: I'll take one of you first.

Kai: Okay.

Tim: So this is kind of what I was going for.

Kai: Called Old Man- in a Studio. Well, we can get rid of that in AI. All right. Now I'm gonna take a picture of you. All right. You look down and be studious. Okay. We are now gonna- Create a prompt. We will feed our draft prompt into AI to have it clean it up, but we're going to, um, specify subject, the action, the setting, the composition, and the style, and then lighting. So, um, the subject. The subject is gonna be, um, uh, Tim or Kai.

Tim: Tim or Kai, or Tim and Kai?

Kai: Oh, Tim and... Oh, I see what you're doing. Okay.

Tim: Kai's trying to write me out of the prompt already.

Kai: Ignore Tim. Um, okay. So, so we have the subjects, which are Tim and Kai. The action is headshots. Um, the setting is the podcast studio. The composition, let's try just side by side just to see. Classic. Yeah, all

Tim: right.

Kai: Um-

Tim: Side by side.

Kai: Okay. What, what kind of style should we use? A '70s retro?

Tim: I like that.

Kai: Okay. Let's

Tim: do that.

Kai: Okay. Lighting.

Tim: Soft.

Kai: Soft light. So this is what we're going to give to Claude to help us write a prompt. The subjects are Tim and Kai. The action is headshots. The setting, in the podcast studio. The composition is side by side. Style is '70s retro. Lighting, soft light, remove the harsh fluorescent light. Camera angle, slightly from the side. Purpose, thumbnail for the podcast episode on AI image creation.

Tim: Do we wanna add the text AI with Kai on there?

Kai: Sure. Text AI with Kai. Okay. So I'm gonna take this, and I'm gonna go over to Claude. I'm gonna, uh, start a new chat, and, um, I'm gonna say, "Please help create a prompt for Nano Banana."

Tim: So you're having Claude write the prompt for Nano Banana?

Kai: Yes.

Tim: Okay.

Kai: But I'm using everything we just came up with in, in that structure.

Tim: Mm-hmm.

Kai: Um, and it is using the new Fable 5 model within the Claude desktop app.

Tim: Mm-hmm.

Kai: Here's the prompt it came up with. Side-by-side headshots of two men, Ca- Kai and Tim, in a cozy podcast studio framed from the chest up. The camera angle is slightly off to the side, not straight up. Given the- giving the shot a candid editorial feel, both men face the camera with relaxed, confident expressions, seated in front of podcast microphones or boom arms. The studio background shows warm wood paneling, acoustic foam tiles, and vintage audio equipment softly blurred behind them. The whole, whole image has a 1970s retro aesthetic, warm amber and burnt orange color gradient, slight film grain, muted earth tones, and a subtle vignette-like vintage magazine photo. Lighting is soft and diffused, like a large soft box or window light, with gentle shadows and warm skin tones. No harsh fluorescent or overhead lighting. Bold retro text across the s- top reads AI with Kai in a chunky 1970s display font with rounded letter forms colored in cream and orange to match the palette. Composition leaves clean space for the text. 16 by 9 aspect ratio, high detail, shallow depth of field. So it added- Wow ... a bunch of things that we did not- Yeah ... specify. So what we'll do now is we are gonna paste this prompt into... First we're gonna do into ChatGPT. Uh, we'll do 16 by 9, then we'll do an, a square.

Tim: Got it.

Kai: Okay? Um, and then we'll also go over to Gemini, and we'll hit the plus button. We'll say create an image, and we'll paste the exact same thing All right, listeners, it is back, and let's, let's have a look here. Well, if you've ever seen the show, That '70s Show- ... you will recognize this graphic. So I would say out of the two, ChatGPT is the winner. Um, Midjourney has proven to be very difficult. Um, I would say NanoBanana did a good job, but w- we don't really look like ourselves, even though we gave it photos. I look more like myself. Tim doesn't look anything like himself, um, whi- which is weird, and there's a lot of equipment behind us. ChatGPT did a very good job.

Tim: Wow, that's kind of cool.

Kai: Yeah. It looks like us. Basically the same styling, same, same type of font, same coloring, and things like that. It's a little more realistic. There's not as much equipment in our background. Um, and it's definitely, I think, the one we would go with. Okay. Yeah. All right. So-

Tim: That's cool.

Kai: I mean, it did a really good job- Yeah ... I have to say. Yeah. Well, folks, that's gonna do it for today. I, I hope everyone, uh, got something out of today's, uh, deep dive into AI image creation. If this was useful for you, if you learned something, uh, you can find more of it on YouTube at AI with Kai Podcast. So go check that out. Hit follow on the show if you haven't yet, and if you've got a friend who runs a small business or is just curious about AI, send them our way. We'd appreciate it.

Tim: And don't forget to check out our AI-generated thumbnail for this episode. Let us know what you think.

Kai: All right. Take care, everybody.

Narrator: For more episodes, practical tools, and business insights, visit aiwithkai.com. Follow the show wherever you listen to podcasts, and share it with someone building a business on the Gulf Coast