GOOGLE'S NANO BANANA: THE VIRAL TREND THAT DETHRONED CHATGPT

E82 GOOGLE'S NANO BANANA_ THE VIRAL TREND THAT DETHRONED CHATGPT

Transcript

Dalton Anderson (00:02.19) Welcome to VentureStep Podcast, where we discuss entrepreneurship, industry trends, and the occasional book review. You've probably seen it on your social media over the weekend, or the last week or so. 3D figurines, these animals being turning into plushies, all sorts of imaginative approaches on life and people. But there's typically a trend of these two.

turning your pet into a plushie or turning a friend or your partner into this 3D figurine, almost into an action figure collectible.

That viral feature has pushed.

Dalton Anderson (00:50.262) Gemini to the number one app, most popular app at the Apple store, not Android, iOS.

Let me repeat that. Jim and I is now the number one most popular app given this feature. Everybody's downloading it for the Nano Banana. The Nano Banana. And I think it's cool that the Nano Banana name stuck, because that's not what it's called. We'll get into a little bit more of that in the episode. Today we're going to be breaking down the Nano Banana, what it is, how it's working, and, well, I'm going isolate how it's working. I'm not going to talk about the models used.

That might be a separate conversation. can read the paper together, but we're going to be talking about how it's used, how it's different than imagine, imagine, and then also how it only does images and some examples. I have not tried these examples. These are legitimate real time examples that I, have made examples before the video, but I haven't tried them.

I've got the photos or they may or may not work. Who knows? We're going to do it live. So let's drive into the nano phenomenon. So nano banana was a code name at Google and Google typically has these code names. mean, a lot of companies do like these big companies. They want to prevent people from leaking stuff, but Google's code name in this instance got leaked.

and it stuck. People appreciated Nano Banana more than...

Dalton Anderson (02:36.46) Gemini 2.5 Flash. And so that's really what's stuck versus calling it what it is. We're calling it Nano Banana. I just think it's funny that the whole point of a code name was so we don't leak that, hey, we're working on a fast version.

of our image and model. It's the Gemini 2.5 flash. So instead of it calling that internally, because it's going to get leaked everyone, it's going to get leaked. We're going to call it Nano Banana. So no one's going to know. No one's going to know what it is. And then it got leaked. And then the name was more popular than the actual, I guess the moniker was more popular than the official name of the product that they're rolling out with. It's kind of cool because

doesn't really make sense that the code name got leaked. And then that's what became the popular name for the model. doesn't go with the branding. doesn't, it doesn't do anything well with Gemini's brandy. Like there's Gemini flash, Gemini pro and it's like 2.5 pro flash. And then there, there's the Gemma models and these other things. And then there's

Then there is Nano Banana.

Dalton Anderson (04:12.534) Okay. So Nano Banana went viral from these two pieces of turning their friends into the 3D figurines and your dogs into these plushy toys. And we'll try out some examples ourselves later on the episode just to see how it works. I hope it works the way that I would anticipate it. We'll see. I'm personally not on TikTok, nor do I spend a lot of time on social media apps besides

X to look at tech news. So I didn't know of this trend, but I did read about the trend itself and saw some images of the trend simply because people were talking about the app becoming number one in the app store and people couldn't believe it. And so Nano Banana did a really good job of this grassroots marketing within

a small cohort on social media, especially tick tock. So I guess it's teenagers and, kids, which is a great clientele to get hooked on your, on your nano banana model, and then do all their schoolwork through your AI. And then they're just, they're addicted to using your tool for the rest of their lives. Like Google that's, that's Google's dream right there. That's Google's dream. So

NanoBanana went viral, right? But what makes this AI model different than their previous approach, ImageGen?

ImageGene had been around for a while and I've used it sometime. One thing that I hate, I was going to say hate, but I don't really hate it. It's just inconvenient. It takes a long time to process an image because it's specialized in these high quality, these high quality images that a lot of people just don't need, right? Like fundamentally you don't need an ultra HD

Dalton Anderson (06:23.598) 12K TV, 1080P does just fine, or maybe you want 4K, but you don't need this crazy upscaled image on every image. so ImageGen was really pushing the boundaries of what it would be for image generation. And it was a, you know, a model first of its kind. And there's just a lot of good things that ImageGen was, right? But as things became more popular,

you could process images faster. There's better models, better ways of doing things. And they've learned a lot, right?

And they've also gotten data from users, right? That's another thing. Because previously, it's just like the tech nerds and the researchers that are using these tools and they find them interesting.

But when ImageGen came out or like the first versions of ChatGPT, Sora, these other things like that, no one's really using them in a practical sense. They're more of like experimentations and admit, you know, I was gonna say adoptions of imagination. And there's just these cool niche tools that were created through

Enthusiasts. mean, long term, there is a thought to move that into a commercialized setting or a setting that will produce potential revenue, but it's far away from that and it has been for a while. So there's just there's just not mainstream adoption of these tools, but that's since changed, right? There's a lot of interest in AI tools. ImageGen, as I talked about on the last episode with Luke Tatman,

Dalton Anderson (08:18.647) is more than half of the companies in the US are looking at image, not image gen, but generative AI.

Dalton Anderson (08:30.671) for their company. And if half the companies are looking at Gen.ai, that also means that people at those companies are using it or know of it. And I also talked about in the last episode that quite a few individuals have used or downloaded or partaken in AI, these AI chatbots.

a significant amount. I don't recall the numbers off the top of my head. I talked about it in the last episode near the beginning. So I have reminded myself before I recorded this one. I say all of that to say this. AI is no longer this niche thing. It's becoming more mainstream. That's clear. So they've got legitimate user data. This isn't the data of some enthusiast or

some researcher, this is people that want to actually get stuff done and they want to do useful things and they, they're not experimenting. They're there for purpose. Whereas like two years ago when I was using it, I didn't really care what I got. I was just cool to be there. I was just like, Hey man, this is really cool. I'm glad you have me here. Whereas other people are going to the bar and they're trying to get a special meal or drink that they really want where I'm just there to people watch. Right?

There's people that are purpose built going to the app, trying to generate something useful. Whereas previously people didn't necessarily care. They were just there to have fun. So this is a different clientele and with a different clientele, there's different expectations and the expectation from these users is it needs to be fast and it needs to do the things I say it does. And then if I need to make changes, I could make changes on the fly.

That's way different than ImageGen, because ImageGen was something that I talked about earlier, where it was upscaled, high quality images, whereas Gemini 2.5 Flash Image creates images pretty quickly, and it isn't too worried about these crazy high quality images. They're quality enough, like believe me. Some of them I saw.

Dalton Anderson (10:58.253) I was like, wow, that's incredible. I didn't know that was AI. They got me. So yeah, I just, I just wanted to call out that ImageGen and NanoBanana are two different approaches fundamentally because there were two different clientels and two different periods of the marketplace. Now we're in the marketplace of users demanding useful stuff and it

having higher expectations, whereas before it was more enthusiast. So I that was important to add.

So I guess let's talk examples and then I can maybe talk about impact and whether it's hype or not hype, right? I don't want it to be discussing all the impact of this AI tool and it doesn't even do the things that we're talking about. So let's try it out. So let me get this tool ready. I'll share my screen.

Okay, so the tool's going. I'll share screen. Gemini. Okay, so I have some prompts ready. The first one is the favor, the 3D figurine. So I've gotta give it an image, plus my prompt. So we'll see how this goes. So I'm turn myself into a 3D figurine. Trust me, I don't need this to be all about me.

it was just more convenient to do photos of myself.

Dalton Anderson (12:37.941) simply because I didn't give people's permission to add them into the podcast. And I have friends, but also I want to be conscious of adding them to this content without talking to them. And I'm in a rush. I want to get this done.

Not in a bad way, but you know, doesn't add that much more value for me to add another person. Oh wow. That's actually pretty good. Okay. So it's, let me describe this photo to people that are listening and not watching the video. And if you like watching the video, you can look at that on Spotify or YouTube. But okay. So there's an image of me. I just got done with like a 14 mile run and

I'm at this park and I've got blue running tights on. I've got white socks, white shoes, my Garmin watch, a soccer jersey on that I got in Mexico city. If you know the club, let me know, say it in the comments. And then I've got my running vest on and a headband. I'm, there's a couple of people in the photos, but I'm the main subject. There's trees that

are changing colors simply because of the way the light is reflecting off of the leaves. And then there's a streak of light crossing through the middle of the photo and piercing my shoulder. So there's a lot going on in the photo and it couldn't misinterpret some of the other people in the photo, but it interpreted me as the main subject, which was correct. And then I had asked it, the photo of the...

Take the person in this photo and turn them into a high quality image, 1 7th scale resin figurine, place it in a collector's shelf next to other anime figures with dramatic studio lighting. And this is me. It looks, I mean, it is crazy how much it looks like me. Maybe my only commentary is my eyes look like that.

Dalton Anderson (14:55.76) Like my eyes in this one look a bit odd. But it looks pretty close to me. Yeah, I don't really have any.

Cause if it's zoomed out, the photo is pretty far away. So it's picking up, it's picking up my facial features pretty well. And it even picked up the strings on the chest where I have from my running vest, two strings are running down the middle. And it picked up that those are, those look to be, I can't really, I guess I can't zoom in. That's unfortunate. me just open this back up. It picked up that these are two separate strings.

Pretty interesting.

It did a really good job with the vest. And there's the photo in general, because the photo is far away. So I'm pretty impressed.

No joke. All right, well, it's a good start. So let's keep on trying to break the model. I'm going to just, just for freshness and to have a control in place, I'm going to make a new track every time just to make sure maybe the chats might mess it up. You can never be too careful. So this one is transfer the dog in the photo into a felted wool.

Dalton Anderson (16:22.608) plush toy, give it an oversized stitched black eye and a slightly smiling expression, placing it a clean white background. Wow, I to really work on my reading skills. Oh, I deleted the photo on accident.

Okay, so this is Lulu. I wish I could click on it, but she's such a cutie. She's a Sharpay and she is our family dog. And this is a photo of her sneaking up on this chair while the house is under construction because she wanted to look outside, but she couldn't look outside because there was no place for her to go. Previously, there's a nice chair for her to look out and watch outside.

So let's do this. And she's on a chair, the chair's similar color to her body and there's only one real subject in the photo so it's easy to describe and understand. Or I would just say decipher, but decipher and understand what's going on. I would just think that there may be some overlap with the chair because the chair's the same color but we'll see how it's handled. So add tools, create images, NanaBanana.

No banana.

Dalton Anderson (17:43.31) And the images are created pretty quickly.

Yeah.

I think that's pretty good.

I think that's a bang up job right there. I mean that's.

Dalton Anderson (18:05.829) That's pretty good.

I think that looks like Lulu. Did a pretty good job with making that look like Lulu. Such a cutie.

Lulu. Lulu is so cute. So, okay, so it rocked that one. So we're two for two right now. We're two for two. So let's try another one. The next one is gonna be, take a photo of somebody and.

Dalton Anderson (18:46.481) Thanks, it's asking me to try their video model now. Well, let's take a photo of somebody and turn it into a cyber.

So I was gonna use the coffee shop guy. I took a photo of a random man in a coffee shop and it was a really dope photo. So let's see what happens. And these are quite quick. All right, I'll transfer in the person in the photo.

did I mess up and not say tools? want images. Take.

where.

Dalton Anderson (19:30.513) Alright let's try that again, sorry. What's a really cool photo honestly? I think this is like cyberpunk lighting. Yeah.

Dalton Anderson (19:46.321) Sounds good.

Dalton Anderson (19:49.968) Okay.

Dalton Anderson (19:54.13) So let's see how that one goes. Once again, it's super fast. Surprised how fast it is. I've used it before, but I was doing things, I wasn't really watching it and coming back.

Dalton Anderson (20:12.625) This one doesn't really hit that well. I'd say this is a 0.5. I it does what I asked it to do but it doesn't keep the same vibe of the bar.

Dalton Anderson (20:27.013) And it just kind of did whatever I wanted, which is probably my bad. It's kind of a bad prompt, but that's all right. Let me just try one. Let's try another one. So recreate a scene of this photo using slow motion figure and creatures.

Dalton Anderson (20:44.565) I don't know, let's try this. That sounds crazy. So let's try that.

Dalton Anderson (20:54.915) So I was at Washington Park in New York City and there was a dude in the fountain. And apparently in the morning time you can go there if you want to see people, I guess, shower slash bathe in the fountain you can. But the water is just like this dark green color because it's the water is just dirty.

And there's trash and stuff. There's not necessarily that much trash. Maybe that's not trash. Maybe that's an air bubble. But I mean, the water is not clean. I would definitely not go in the water. What? This image is crazy. It did something weird with my eyes. Like the AI model doesn't like my eyes. So yeah, this did not work. So I'll actually count this one as like this did not do what it's supposed to do.

but it also isn't a single person. So I could try with that 3D photo and see if that works.

Dalton Anderson (22:04.353) see but yeah I really mess with my eyes. I mean my eyes are all funky.

It didn't do anything besides remove the man in the fountain, or crop the photo, and mess up my eyeballs.

Dalton Anderson (22:24.645) Yeah, I don't think this prompt is working. The prompt was recreate the scene in this photo using stop motion clay, clay, me on figures. I butchered that. Give the characters visible thumb prints in the clay, slightly imperfect, handmade look, put them on a miniature set. That was the.

prompt, so let me just make it something easier to do. And then we'll talk Biz after that. I have some modeling examples, but I also wanted to try something that people would actually use it for, right? Like people aren't going on here to do a lot of, like the main users right now, like are not business users, they're.

They're just people, right? So I'm doing funky stuff that a person would do to mimic what would actually happen.

Dalton Anderson (23:34.577) Yeah, this didn't work either. Turn the person into this photo into a 16-bit art character from the classic SNES role-playing game.

I would say nothing happened.

Dalton Anderson (23:58.406) I don't know, it seems fundamentally not to be working.

We were two for two and I was very surprised.

Dalton Anderson (24:12.689) And now we are to...

Maybe it did some weird stuff to my face. Maybe I have to get like a close up.

Dalton Anderson (24:27.51) Maybe I'll try Luluh, because that's a close up. So let's try that.

That's a close up. Turn this person.

Dalton Anderson (24:39.611) Change from person to dog.

Dalton Anderson (24:49.777) I'm being a little stubborn here. I'm trying to get this to work.

Dalton Anderson (24:58.607) Yeah, it's not working. Okay, so I call it two for two. I was so impressed with the first two and I was like, what the heck, this is crazy good. But hey, didn't work out. So now let's try something that I also heard that people are using it for. So I had an image of a model.

that is standing in like a nicely lit studio. It's got a nice background. Wearing a shirt and jeans, got his hands in the pocket. So now I'm gonna say using the provided image, change the color of the model's shirt to navy blue. Let's see if that works.

Dalton Anderson (25:48.785) I was also messing around with this. Let's see.

Dalton Anderson (26:00.146) So the drum roll, drum roll, and it works.

Dalton Anderson (26:08.046) It works.

Wow. I mean, it does it pretty well.

preserve the image in every regard. It's literally identical. I can't tell. I'm trying to click back and forth to see if there's any weird stuff going on in the background or if the light is reflecting weird on the guy's shoes or if the veins are different on their arms or if the hair is a little different, but did a really good job of, I mean, that's really good. Okay.

So now let's change the shirt.

Let's keep the shirt as is. Change the pattern of the shirt to floral while maintaining the original fit and lighting. So now we're going to keep in the same chat and just see if we can keep making image changes to an image and get something completely different.

Dalton Anderson (27:10.381) Once again, never done this live, never done it with the new model.

Dalton Anderson (27:18.043) Wow.

That's so sick. That is so sick.

That is crazy. So we went from a normal photo of a model, white shirt, hands tucked in pants.

Dalton Anderson (27:40.828) to changing the shirt from white to navy, and then we went from navy to a navy floral pattern, and the floral pattern consists of a navy blue base and red and yellow flowers. And that's a sweet, sweet shirt. Okay, so the next thing that we'll do, it'll say replace the shirt the model is wearing with a black,

leather biker jacket, keep the model same pose, face, and background identical.

Now we're just completely changing the whole vibe. I wonder if this works.

Dalton Anderson (28:29.131) Yeah

I mean, that's...

That's crazy.

It's identical.

Dalton Anderson (28:45.356) It is identical.

Wow, wow. Let's ask one more question. Let's say, I'll use the microphone. Okay, looks great. Can we keep the same model, take the biker jacket off and then use the floral shirt that you provided earlier and then have the model doing something with its face, like a pose that the model touches their face.

Dalton Anderson (29:20.699) We'll see, mean, that's a hard one. That's a lot going on.

Dalton Anderson (29:31.991) No way! What?

Dalton Anderson (29:38.927) That's insane! That's crazy! my goodness. my goodness.

Dalton Anderson (29:54.865) Speechless. I'm honestly speechless that that was done live and flawless. Insane. Let's just go back and run it from the top. So this is the first image that you'll see. White shirt, hands tucked in, black shoes. It's kind of hard to tell the color of the pants given the background and the lighting, but it looks like this like greenish gray slash black color.

changed the shirt to blue, navy blue, which looks really nice. Then I changed the shirt to this navy floral color. And then I changed this weird biker outfit, which was odd. Doesn't fit the outfit at all. But then I'm like, hold on, I made a mistake. Let's just keep the floral outfit and do a hand on face gesture.

Dalton Anderson (31:02.395) crazy.

Dalton Anderson (31:07.09) I'm doing the same gesture, I don't look as cool, but. I am so impressed.