Nano Banana Pro can do what??

Hey there, my gorgeous friends on the internet. We got a new big juicy banana in town as some call it Nano. I don't think it is though. I've been playing around with Nano Banana Pro for the last couple of days just to kind of see where we are right now. Uh the past year or two, you know, when I tried these image generation models from MidJourney to, you know, GPT image, the novelty wore off really quickly and I wasn't really impressed with the results. But with Nano Banana Pro, I feel like we're getting to a level where oo maybe we have a potential here to use this for different SAS products. So I want to kind of show you what my results are. So the first test I want to run is essentially convert a 3D model here uh into pixel art. And the idea is that if it does really good job at replicating it in pixel art, then it's really easy to kind of move a 3D model around in positions you want. So you can maybe have the frames that you want to extract from it and then just generate that to create maybe like a sprite sheet. And to my surprise, wow, this did actually a fantastic job. Have a look at that. So this is the original 3D model which is just exported as a PNG. And this is what it created from it. And that is pretty much on point here. uh from the the necklace there or what do you call that the choker uh to the socks and the shoes, it's pretty much pixel perfect. So really really impressive. However, I thought to myself, would we be able to just provide one image here and then kind of generate the sprite sheet? So here's my prompt. Create an attack sprite sheet out of character where they swing the cactus. The spreadsheet should be all horizontal with equal spacing and transparent background. One of the big problems is that it has no idea how to get a transparent background. Right now, I also told it to do equal spacing and all horizontal uh which it did not get. So, that's a shame. That's like one of my main issues right now still with uh these image models is that they cannot do proper transparent background. And for this use case, it would be actually really fantastic. Um, and then like not having the equal spacing here, it would be really hard to generate an animation out of it. But for the actual like images that it generated, they are pretty good. I mean, the image is pretty much the same. So actually like there's no like the character doesn't start looking differently, which is really nice to see. And the way the hair moves as well is really impressive. and just kind of the animation how it flows looks really good. So she's turned over to the left side. That's that's that's a little unfortunate, but the way she swings this, right? So that's one pose. And then here you even have a little blur here added to it, which is super cool. And that's the finishing position. And then back to normal. So it actually gives you quite a bit to work with, which is really nice. Or you can also use this as a reference um to kind of recreate it yourself. But pretty impressed with it. I tried this a couple of times, but I just couldn't get it to be consistent when it comes to the spacing. Uh, so this is another character I gave it and I told it, "Hey, create a walking animation sprite sheet out of this character with transparent background." And again, background's not transparent. It's a JPEG image, unfortunately. And as you can see here, the spacing again is is quite off. However, I did manage to cut this out just to kind of see how the animation would look like and the walking animation is unfortunately not nice in this one. So, as you can see, this is the result you would get right now because the images are not equal in spacing. Uh, but if you just look at the actual animation is not too too good. And just as a reference, I actually have a walking animation that's done properly by someone. And as you can see, the difference there is quite astounding. Like you got the hair moving, you got the sword kind of wibbling and wobbling around. I'm just really hoping that in the future we will get to a point where you can just feed it maybe one or two frames of a character and have it make full walking animation or attack animation. Wow, that would be so exciting. I'd actually love to make a couple of games. If you guys don't know, me and my wife and my baby are in the process of getting our first home and it's really exciting. And I actually found Gemini being really, really good for this use case. If you want to remove an item to kind of visualize how things look, or if you want to redesign a room, wow, that's such a fantastic use case. So, for this example, this is one of the rooms uh upstairs, and we're like, get rid of this whole pink [ __ ] Like, I don't want to see any of that. How would that look without it? So, that was the prompt in GPT. And look at that. That is is so so good. And like older models, you'd always have this problem where it would actually just like regenerate the whole image and change like how the bed looks, how the books look here on the right side. Uh, but Gemini Gemini Nano Banana Pro does a really good job of just targeting specific parts of the image, which I really, really like. So, look at that. How good is that? That's really impressive. It even got rid of the glare here. Uh, another use case was to redesign something. So this is how the dining area looks like. Horrible. Absolutely horrible. So I prompted it to essentially add a modern twist to it. Maybe a contemporary design look. And this is what we got. So have a look at that. That's pretty good. So again, it does a really good job of kind of keeping everything that's part of the original image. As you can see, you have the plugs here on the bottom right. Uh what else? See the chimney there as well? You have that. So, it keeps everything really, really nice. You know what else is really cool? Being a problem solver. And that's where today's sponsor, Brilliant, comes in. Brilliant is an amazing learning platform that unlocks that part of your brain that just gives you that confidence and that intuition to pick up any new skill from math to computer science to large language models. And you can do that by learning through these nice visual, highly interactive problems rather than passively learning through videos. And what I love is that you are actually solving all of these problems. And Brilliant is there to just guide you to those aha moments where things finally click together. One of my favorite courses that they have is the thinking in code. And it doesn't teach you just syntax. It teaches you how to think like a programmer, how to structure logic, how to break down big problems into smaller ones, and also build intuition that you can use anywhere, whether that's like in Python or GameDev, or just being faster at solving tough challenges. So, to get started, head over to brilliant.org/developed, or you can scan the QR code right here, or click the link in the description down below. They will also give you 20% off the annual subscription. So, you have access to all the courses on the platform and all the lessons unlimited daily use. It's fantastic. So, thank you so much and let's get back into it. I also need to create thumbnails for every video, but I wanted to see how far can I keep making changes to an image before it just completely falls apart. And I've noticed actually if you prompt it a couple of times, it just goes haywire. I know a lot of people say you can just ask it, hey, change this little part, this little part. Uh, but I didn't have the same experience. I I found it really hard to uh to keep prompting the same image and keep consistency all across. So, this one was create a YouTube thumbnail out of this, cut the person out, and make the background showcase Gemini tree pro capabilities. I told it basically to do a doodle here and then a really like high realistic image. So, I gave it myself as the image. did did a great job cutting cutting me out and it did the doodle and then the other thing there. Uh but I didn't like the background. So I was like, well, get rid of the background and just make it all dark. Uh which it did. So so far it's done a great job. But then I said, you can move me a bit to the right side and make the doodle a bit bigger and the realistic one as well a bit bigger. I don't want the black background to take up much space. So kind of did it. But then I was like I changed my mind. So, I said, "Let's make it a bit nicer. How about position me in the middle?" Uh, and make myself smaller. So, kind of here. And we can literally have it split in the middle. Kind of like Street Fighter versus screen, right? How you have the two characters. So, the doodle and the realistic image should be the same size. And it just kind of didn't do it. It It pretty much output the same image. So, then I told it just zoom in and remove the background. Essentially, have the doodle and realistic covered. take the whole thumbnail and make myself a bit bigger, too. Right. So, it just get rid of this whole gray area here and just make the doodle full screen. And then it did the it basically did that, but it got rid of me. Which why? And then I said, "Okay, perfect. But you removed me, so put me back in the middle." Who the [ __ ] is this guy? Why is he praying? I wanted to see if we improved in the logo creation category because I never liked any models pretty much when it came to generating nice logos and surprisingly we have improved on that front but not if you just simply prompt it uh like this for example. So I said create four logo variations. The logo is a semicolon. It should be coding related. So, it did that, but it still looks super AI generated to me. Where I see Nano Banana really shine is that you can get give a couple of reference images and kind of basically combine multiple different images to kind of get the results that you want. And it does a really good job with that. So, if I add myself, for example, and maybe some clothing, it does a really good job of taking me out and putting me in that context with the clothes on, uh, or in maybe a specific area. Um, but if you just prompt it like this, not too good of a job. So, for example, here, see, I gave myself as a image doing the tea pose and then this uh sweater here, and it does a really good job of putting the sweater on me and keeping everything else kind of the same way. So, this is how you should use it when it comes to making logos as well. So, I found these logos. I'm not sure who it was by. I think by Open AI, but I really liked how they looked like. So, I said, "Okay, use this kind of as a reference and generate me new ideas for logos based on that semicolon." And it did a much better job. Have a look at that. Uh, would I use any of this? No. But it's a good starting point here for sure. I kind of like the idea of this bottom left one. And the top two look all right as well with a bit of work and refinement. You can actually turn it into a nice idea. So much better job if you just kind of provide some sort of reference uh to Nano Banana. Uh the more I tried to to do it, the worse it got though. So I said, "Okay, do it again, but do four more variations." And here we go. Uh I guess that's that's not too bad. Uh but yeah, and then I said, "Do it again, but make it kind of like binary code." And then it did this which looks looks a bit [ __ ] I was wondering what kind of sass idea would you be able to make out of this. I was like, "Oh, what if you do like a upscale and recolorize old mangas?" But then I was like, "Oh [ __ ] do you think they would mind with all the copyright things or no? They would they would probably mind, right?" But did a really good job. So this is from Death Note here. So you can see black and white quite quite low resolution as well. And here is the result that Nano Banana Pro gave me. And as you can see, it pretty much kept everything the way it should. So, it hasn't really like altered or modified the image. Uh, which is super nice. Um, and the text even looks sharper. It has the same kind of font to it as well. So, really, really impressive. Shame again that we probably would not be able to make a product out of this. Uh, but another example I thought was like, oh, what if we could like visualize? It would be really cool to have like a book, right? And you can basically create like a visual comic from every page. So, I was like, oh, let me try Harry Potter and see how that would work. So, what I did was basically give it the first five pages of uh philosopher stone, Harry Potter, right? And I was like, I'll generate me a nice comic style. And here's what we got. It's It's pretty good. I'm not going to lie. I'm quite impressed that what's his name? The big fat dude. Uh he he looks quite actually accurate to the movies as well. Uh but yeah, there you go. That's a cat reading a bap. What is this wizardry going on here? Uh I also told it to do another version of it and I think it was pretty much the same. Yeah, here he's holding a donut. here. He's holding some sort of raccoon for some reason. That's worrying. Since I have a little boy now, I I really wanted to see if I can create some cool infographic cards in a specific style. And I wanted to see if it keeps that consistency, and it did really well, actually. So, let me show you that. So, I said, create an infographic and a flat cardboard style cutout that would explain addition to my kids. I don't know how to spell addition. Uh but that's okay. We have AI for that. Uh so here is the first image. So I did addition, right? And did a cardboard cutout. Pretty cool. Done really well. I did apples. So we have three, right? We know that we add two more. Plus means putting together, right? And then we have five. So it did a really good job. We have five. But it kind of messed up the coloring a bit. Now that's confusing to me. We have two here that we're showing that we're adding highlighted in orange, but when we add it, it's three. Unbelievable. Unbelievable. Should have been too orange here. Silly. Uh, but it did. Okay. So, I said the same thing, but do it with subtraction. And yeah, it kept the same style. So, again, really nice and consistent. Uh but it did the same thing where it started highlighting this as orange here at the end. But other than that, pretty good job. Now the most confusing part is when I asked it to do multiplication. Have a look at this thing. What a mess. Um so we got two right times two. What is this picture here supposed to mean? Are you trying to confuse my child? Shame on you. We have two, two, and then you do the multiply and you're showing me three. What's wrong with you? Get help. And then here at the end, we do have four. So, I did end up doing it properly. But again, the way I highlighted the color here is quite random. Uh, but close close enough. So, maybe the next version we'll get there and we can actually create really cool infographics. I tried it with other stuff as well like uh with like code like show me how use effect works and it did did an all right job. Uh nothing too too crazy or impressive. I'm not too big of a fan with how this looks like but it did okay. Right. We got a component mounting and rendering here. It shows you when the effect runs and then if we have a dependency array then it runs once and then it stops. If we have other dependencies, it checks if the dependencies have changed. If they haven't changed, yes. And then it goes here. Blah blah blah. Right? We know. And again, I feel like if we managed to get the whole cutout thing sorted, so having actual transparency, we'd be able to generate some really cool uh like images for, for example, a weather app. What a great use case here. So imagine you put in the city that you live in, right? This is Abberine. This is where I'm currently at. And it'll generate you like a cool 3D isomorphic uh background image here. And so there we go. So I tried that. That is Aberdine. If you're interested in the prompt, it it's something like this. So present uh clear 45 degree top down view vertical isometric miniature 3D blah blah blah. Right? And then at the end here, you just add silly name and then you add the name. So you could just have this in your back end, right? It would generate the image for you. And it's really good and it's really consistent again. So you have this right it's Aberdine. This is another city here open in Scotland and again it kept the same style. Really fantastic job. And then for another example here we have London as well. Have a look at this. So yeah really cool. It's I'm honestly impressed with this. Again if we can just extract have this as a PNG. This would be juicy. Look at that. look really consistent. I'm honestly really impressed. Okay, I've been messing around a bit more with this and I think I found a way that this could work in a way. Um, I am planning on streaming on Twitch in the future and I was like, well, let me see how I would be able to generate like emojis, right? Uh, so I provided an image of myself and I was like, give me some emojis in sticker style. And I said, actually do the the background all in green. So you could actually do it like this where you just kind of have it like green screen and then you can use that to essentially be able to cut it out either yourself or even in JavaScript you can you can write a little uh like automatic like keyer and that will get rid of the green for you and it's pretty good. Would I use this? No. Sh. But uh it works right. It does a really good job. But yeah overall I'm really impressed with the banana app here. Uh, I I did one more based on this pixelated character. I said, "Make a a League of Legends champion out of it and see how that looks." And it actually did a really good job. It's it's pretty much that same style. Uh, so super super cool. Uh, it's going to be interesting to see kind of the use cases uh, for this. I'm definitely going to try out a couple of things. But yeah, it's really nice that we are to a point where we have that quality now and it's not some like low low res crappy image and we also are able to have that consistency where we can add two images or three images and kind of keep reusing them without them starting to look like a different person. Uh so yeah, really exciting. Let me know what you think. Uh if you want me to prompt other stuff, leave it down in the comments below and I will do so. All right, catch you in the next one.