Nano Banana 2 is Here! Full Review & Testing Results

So we had another big announcement out of Google. They upgraded their best image model. And here's the thing. Many people already considered their previous image model, Nanobanana 1 and Nanobanana 1 Pro, to be the best in the industry. Sure, some opinions say that maybe Midjourney is the more tasteful model, but that's pretty much it. Other than that, people really love Nanobanana 1 and Nanobanana 1 Pro. And now we have Nanobanana 2. So I thought this would be worth a separate video where we compare the outputs to all other popular image models out there. We verify a bunch of the very impressive examples they showed off on their launch blog post, and we run the prompts ourselves. And 3, we show you a bunch of use cases for useful stuff that the previous model wasn't able to do as well as this one in a little live section with a bunch of prompts that you can take away and use yourself after this video. A lot of ground to cover. And honestly, I like making these image generation videos because they're so visual and it's so easy to see AI's progress because it just looks better and works better rather than LLMs where a lot of times it's kind of hard to pinpoint which part of the text is now better or different. Let's do this. So here we are. If you're familiar with the channel, we have a few prompts in different image generation categories, and we just compare it to some of the most popular models. So here on the left side, you see the Nanobanana 2 output to this logo design prompt. And then on the right side, now you see Midjourney, you see the Grok image model, you see the ChatGPT image model, there's Flux 2, and then finally Nanobanana Pro, which was the previous best version. If you want the full sheet with all the Chinese models and previous versions, we'll link it below. I would say on logo generation, it's not special. That's not the biggest strength of this model. But now we go to portrait photography, and I think here it actually really shines. You know, we had really realistic models before. Even if I look at Nanobanana Pro, it looked really good. To me, this looks even better. Maybe it's kind of the textures, the pores on the man's face, or it's the posing. I don't know, it just looks perfect to me. Next one is a cinematic still, which always is an interesting one because there's a lot of variety in what you get from the different models. I like it. It's a hyper-realistic movie scene. This could be a frame from a movie, and it's not that stylized. You know, when I look at some of the other ones, including Nanobanana Pro, it uses a lot of saturation and a lot of style. And this one is like rather neutral, just real. And I think that's a thing that remains true across the models testing aerial photography, kind of all outputs look very similar from the various models. So this one is not that exciting. But yeah, it looks good. It looks very photorealistic. It's incredible at that. And then when it comes to book cover, also a lot of these look similar lately, I noticed. It's really good at graphic design and book covers. I would say this one is actually maybe a little better than the previous Nanobanana. It just looks a bit more aesthetic, the whole composition and the way the colors match. And then lastly, this one was interesting. We asked for a comic strip illustration and it actually refused because it figured like, hey, maybe there's some copyright content. So that's the thing with this model. It's quite restricted in terms of what you're allowed to do. Oftentimes you'll also upload like an image of yourself. You want to remix it and it'll be like, no, I can't because like you're remixing a celebrity. And now just because you're thinking like, okay, yeah, Igor, you do YouTube. So maybe your face is out there. No, this happens for various people. And the guidelines are just so strict that it just cuts you off sometimes. Often the key to resolving that is just running it a few more times and sort of it just works. Anyway, so that's our little comparison there. I think except of maybe logo design, to me, it's a no-brainer to just make this my go-to model for whenever I need image generation stuff, just because it looks so real, it's so precise. But now let's look at some of the examples from their blog, because they led with various characteristics that were really beyond anything we've seen so far, like not a head and shoulders above anything else, but like further. So here's one of the examples they shared in their blog post, like this beautiful top-down infographic, but in this like handmade style. What would you call this? There's like a word for this. What's it called? Various stuff, like collage. Scrapbooking. Scrapbooking. That was it. Yeah. Oh my God. Nice. Human 1, ChatGPT 0. Like a scrapbooking style. I think this is gorgeous. We ran the same prompt that they shared in the blog post just to check how much they cherry-picked the results. And this is what we got. It's still good, but it's not that good. So yeah, as per usual in the blog post, you know, they run it a lot of times and they pick the best one. But when it comes to like the accuracy of all the elements, like it's all there. I guess you could be more specific on asking for the scrapbooking style. I just really wanted to show this too, because I think this is a very creative way to add like images or B-roll to your video content or social media content. Like people would not associate this with AI-generated images, right? Okay, the next one is this infographic that is more in this cartoony style. It's an overview of 3 different types of clouds. We ran the same prompt that they shared and we got this. Way more details. And I have to say, I'm kind of impressed by the text. Usually, like, you read through it and you find a letter off here and there. Text got so much better over the last year. But here, I kind of have a hard time finding a mistake. I think all of this text is just flawless. Correct me if I'm wrong, though. Even the numbers, it's just good. To me, it seems like it went from, like, 95% reliability to at least very close to 100%, right? Okay, here's the next one where they compared Nanobanana to Nanobanana Pro. Nice. We ran the same prompt and got this. I personally like this one more. But it is a different style, but it is beautiful. Like the amount of details and the composition and everything. Like heck, if you're gonna do something in this direction of like infographic-y or like handmade scrapbook-y, like this model is just incredible. And that's also what they claim. They say that the visual fidelity has been upgraded. And I honestly see it. They claim to have more vibrant lighting, richer textures, sharper details, maintaining high quality aesthetics at the speed expected from Flash, their quick model. A lot of that is subjective. You can make up your own mind based on examples I showed you. And hey, by the way, if you're enjoying this video, make sure to subscribe to the channel. We make deep dive tutorials and news updates and short tutorials every single week. And now let's continue with the video. Next, let's look at the stuff that is not so subjective, namely the upgrades to subject consistency and how precisely it follows instructions. This is the one that is really exciting to me because some image models, especially ChatGPT, you tell it certain things and it just ignores them. I was for 3 months in shop for 2 months. We spent, you know, almost half a year together. Well, I have a few prompts here that I want to test with you live to both give you inspiration for what can be done and also test if all of these upgrades are actually present. So let me just go into my Gemini account and click the create image button here at the bottom. And yeah, this automatically has Nanobanana 2. And I'm just going to start with this prompt. Create a bobblehead figurine of me and then attach an image of me. Now, this is not the test yet. What I want to see is if I can bring that figurine into various scenes. Without it turning into a completely different figurine of me. In other words, if I use this figurine in 5 different scenes, is it still gonna look like the same Igor bobblehead that it creates here? Okay, there you go, Igor bobblehead edition. Not bad, that's actually kind of cool. Here comes the real test. Now put the figurine on the table of the US President in the Oval Office. You can see loading Nanobanana 2, and there it is. Okay, let's be super critical here. I mean, the hair is good. Yeah, honestly, I would say 9 out of 10. In previous models, I would say the consistency here was like a 6 out of 10, maybe a 5 out of 10. It's not perfect, perfect. If I'm super critical, I would say that maybe, well, I don't even know, probably just a lighting change. No, this is really good. Let's try one more. Place the figurine in the hand of an astronaut stepping onto Mars for the first time. Heck yeah. Okay. Yeah, that's still the same figurine. The consistency is incredible here. Okay, okay, okay. We like it. We like it. So you can totally do this, you know, create one asset and then put it into various situations. Amazing for, you know, product photography and e-commerce. You can kind of just add one image and all of a sudden, because this stuff stays consistent, finally, you can place it anywhere. No specialty tools, no crazy workflows and apps that take multiple days to learn. Just good old Gemini on a $20 plan that I think you can get for $10. And I think Downabananato even works on a free plan. Yeah, even on the free plan, you could just try this. Amazing. Let's switch it up a little bit and say create 4 more bobblehead figurines of the following characters: Penguin, Cat in a Hat, a turtle, and a lion. And here's the fun thing. They claim that the model can now handle up to 14 objects like this. This is the first time I've ever heard a claim even close to that. 14 objects that it consistently can place into various scenes for you. So if we're gonna have like these 4 figurines plus me as the 5th one, look at them right here. Beautiful little lineup there, especially the Cat with a Hat and Maybe even the penguin. Yeah, I think the penguin is actually my favorite. We prepared a pretty precise prompt here. And this is another thing that I told you about that the model is really good at adhering to details in a prompt. We call this prompt adherence. So when I put in this detailed prompt here, let's have a look. Create a playful beach scene with these 5 characters. Actually, I'm just going to replace this and say these 5 bobblehead figurines. All characters except the penguin are wearing sunglasses. Okay, so we're adding glasses to all of them. Okay. The penguin is holding a tropical drink with a small umbrella in its right hand and sipping from it. And then a bunch more details about the rest. Let's just run this and see if it can actually do all of these detailed instructions, add it to the figurines that exist, and then place them in a completely different scene while maintaining character and detail consistency. All right. Wow. This looks impressive. So editing team, I would love a good checklist on the side. Let's see if all of these things are present. Starting out, playful beach scene with all 5 figurines. Playful beach scene set. 1, 2, 3, 4, 5. Perfect. All characters except the penguin are wearing sunglasses. Check. Not bad. The penguin is holding a tropical drink with a small umbrella in its right hand and sipping from it. Yep, that worked. The turtle is lying on a sunbed, tanning while reading a book. Oh my God. He's totally— yeah, tanning and reading a book. It's perfect. He actually put the book away, but like, it's accurate, right? He's in the process of reading the book. Sure, he could be holding it, but no, that was ambiguity in the prompt. We said he's primarily tanning. So actually, this is super accurate. To the prompt. The cat and the lion are playing beach volleyball together in the sand. Yep, they're definitely playing. Looks like the cat is about to score. Oh my God, this is so good. My bobblehead version of me swimming in the ocean nearby. Yep, right there. Redressed me into a swimsuit as appropriate here. The setting is a sunny tropical beach with blue water, palm trees, and a bright summer atmosphere. Style should be colorful, fun, and slightly cartoonish. Nothing short of perfect. That's really impressive, actually. Like, I've seen so many of these models in action and It's always something that just doesn't perform as you would expect it to. You ask for something, it doesn't show up. You ask for a character to be shown up, all of a sudden they look like a slightly different person. Look at that. Even this cat, the hat is the same. Look at these red highlights and the ribbon here on top. Same thing, red highlights, ribbon here on top. The penguin has a little red bow tie. Look, all these details, they just carried over. This is so freaking good. I guess I found one thing if I were super nitpicky, it would be this turtle seems like a different character, but we can make edits, right? Make sure the turtle is consistent with the original bobblehead. I'm not even sure if this is going to work. I just want to try. Okay, added these boxes, didn't change the turtle. So fair enough, that didn't work. But I'll show you one more. What you could totally do is make edits to, for example, the format. So change the format to fit an Instagram story of this image. I'll add that too. Update. Yeah, there it is. Just extended it. Amazing. You just post this on the story if you want with text up here. So good. Mmm, so good. Now look, I want to say one thing about the prompt adherence. Sometimes it can be a double-edged sword, right? Because it's so precise. Every word that you enter there really matters. So if you don't express yourself precisely, it will have to interpret it. So be precise with your language. Then this is amazing. And in case you don't like something, then go back and change the prompt. So that the words exactly represent what your idea is. And then there's two more fun things that I found on Twitter that I want to show you. One of them is from Kris Kastanova that always experiments a lot with AI image models. Great follow on X, by the way. She created an anime-style panorama of a location. And what I'll do here is just upload this Google Maps image of Midtown Manhattan in New York, somewhere around Central Park here. Let's create an anime-style panorama of this location. Oh, look at that. Not bad. Kind of a good illustration. Honestly, I don't know the street well enough to judge if this is entirely accurate. Nevertheless, it looks great. Kind of a fun one. One more thing that I wanted to show you, which is its ability to replicate handwriting. Can be also kind of scary, right? But let me just show you this. Okay, I'll just scribble a quick sentence. All right, so this is the image. Hey, what's up? I wonder if the font will match. Let's see if it can continue in a font similar to mine. Add text to this picture of a thank you note for moving help. What I'm hoping for is that it's gonna match this font. And extend the writing so that it looks exactly as if I handwrote it. Okay, let's try one more time. That's not perfect, obviously. Match the font color and style. That's way better, but not perfect. You can see it's good at handwriting. If you want to make images of handwritten notes, it's really good at that now. But this is not a perfect match to my handwriting, which honestly is probably a good thing. But yeah, it's really good at handwriting notes as long as you're not trying to perfectly match what you're doing. I guess some of the examples on the internet are a little overblown. On this one because people state that it can match any font. I don't know about that. And that's it. Nanobanana 2. Pretty impressive, right? Several things you can try on social media or to update your website. I think the prompt adherence is incredible and the fact that it can maintain all of these characters. Wow. Really opens up a lot of new opportunities and makes it so much more practical. Product shots for online shops or just telling stories with images that are consistent. Amazing update. And if you found some unique use case for it, please share it in the comments below. That being said, my name is Igor Pogany, and I hope you have a wonderful day.