NANO BANANA TUTORIAL: 16 Prompts & Pro Tips (Google Gemini 2.5 Flash Image)

Key Takeaways

Google's nano banana is not just a thing for fun. Today I'll prove you that it can be useful tool in different cases. For example, I have this render of a room and I want to see how it will look at evening. I just described changes I want to make. And here's it. Result is just perfect and it took only 30 seconds to make or I can change color of the walls without changing everything else or the color of the couches. Of course, I can add and replace objects using some reference images or just using a prompt. The obvious advantage of Nano Banana is that you shouldn't use complex prompts. You can use natural language to describe the changes. Landscape designers probably won't become jobless. But look at this photo of a garden on somebody's backyard with lots of plants. And we can see how this place will look if we add curved swimming pool. You can prototype your ideas of how you can restyle and reshape the place on the uploaded image. I have photo of a backyard with a swimming pool. And I'm going to replace this fence with this wall from the reference image. Result is just nice. But don't forget that when you use multiple reference images, the last image you upload will define the aspect ratio of the generated picture. Okay, I have a photo of a building. I can ask Gemini to make CAD style elevation. And here's it. Nano Banana tries to make such things as accurate as possible. But of course, you cannot be sure that it 100% accurately recreated all the details and proportions. And sometimes it adds little bit random text like in this case. Red roof and concrete foundation are not in places where the arrows point. But there's much more impressive option. You can generate realistic architecture visualization from a single sketch. It can be stylized to look like 3D render for example. So here's a sketch I used as reference and generated visualization. It's extremely accurate as you can see, but you can make it look even more realistic by asking Gemini to create photo instead of render. So here's sketch image that looks like 3D render and photorealistic visualization. Yes, it's still looks a little bit too sterile for a realistic photo, but the light, shadows, and reflections are very natural. Lighting, weather, and time of the day can be whatever you want. I'll add evening sunlight, for example. It still looks great. Materials of the same object can look different on different images. So, you should specify such things in your prompt to get exactly what you want. I can easily edit color and type of the surface of any object. I have photo of a white car and I can make it red or I can change color to iridescent purple. generated image perfectly matches my request. You can also change type of the car surface to matte black. Okay, but it's still glossy enough. So, I'll specify that I want to get 100% matte black car. So, here's previous version and new one. And I also created chromeplated variation in pink Jeep with glitter. If you've ever tried to colorize a monochrome photo in Nano Banana, you probably ended up with a result that still resembles the colorized photos of 100 years ago. This happens because of the word colorize in the prompt. Instead of this, you can ask Gemini to make image that looks like it's made with some modern digital camera. Describe colors of the objects on the image, lighting, etc. In most cases, you'll get better looking image. Nano banana allows you to change color of the hair of the person while keeping everything else the same. It can be whatever. Head position can be little bit different on different generations. Even when you ask Gemini to keep everything the same, of course, haircut can be changed too. You can use prompt or some reference image. In this case, you should clearly explain which image should be edited and which one should be used as reference of the haircut. I specified that the image with freckled girl wearing t-shirt should be recreated, but long hair should be replaced with exact haircut from the reference image. I additionally reminded Gemini to keep the original face the same because in some cases it tries to replace both haircut and face. I didn't find the way to apply makeup from reference image. It always replaces whole face despite I begged Nano Banana not to replace the face completely. If you know how to do this, let me know in the comments. But you can describe things you want to change in the prompt. You can turn your photo into a professional shot from any angle and with any background. You can use such generated images as profile picture for LinkedIn, for example. Nobody will even spot that this is not a real photo or it can be photo set in the same location. But in this case, you'll have to use some upscaler to improve image quality because the quality of the generated images isn't bad but not extremely high. If you search in Google for free upscalers, you'll find different options. Each of them can use different algorithms. For example, I almost always use 4x upscaler in Recraft AI. This is not paid promotion. It allows you to upload and edit up to three images per day. Here's the result. Recraft tries to make images sharp and smooth at the same time. And as soon as we can generate images of any person in different locations and situations, we can make the person on the image wear some clothing or accessories from the reference image. This green jacket replaced jeans jacket from the original image. It shouldn't be only one or two items. You can add much more things at once. Sometimes it works better. Sometimes the patterns on the reference and the generated images may not match. Or, for example, the item can be too big or smaller than you expect. But in some cases, results are 98% perfect. This of course opens up possibilities for product placement. If a person is already holding something in their hands, you can replace that object with your product. But what if you don't have clear vision of how your product should look? For example, you already have logo and maybe some other visual branding elements. You can add them to some existing photo of a box, t-shirt, can, bottle, etc. Or you can upload your logo and generate product images from scratch. You can look how some typography, logo, or any other image will look on almost any object. It can be wall of some building, car, or whatever you want. If you already have images of your product, you can create stunning visuals for an advertisement. Add any effects you want or add the product to any scene. For example, I'll type make this can fly vertically amidst the ocean in the storm. Half drown in the water. Bright evening orange and pink sunlight. Flying coconut halves. Add splash from the can. Look at this product is very consistent. I can use it in social media or some advertisement after upscaling it and fixing colors a little. You can add good-looking effects to some typography when creating posters or just looking up for ideas. For example, I want to make this text look like it's made of glass. Not bad. Let's make this glass a little bit more iridescent. Now I want to make these letters chromeplated and place them in the field full of sunflowers. Looks great. I can also make it liquid. It still looks good. What if I'll try to make these letters look like they're made of ice and ask Gemini to place them into the snowy forest. This is exactly what I wanted to see. I really like how the sunlight is going through the ice. You can see that the colors of the letters from the original image are still slightly visible. So you should specify in your prompts if you want to keep original colors or change them. I just used redo button without changing the prompt. And now the ice is crystal clear. And don't forget that you can change the text while keeping the colors and overall style almost the same. So you can see that it changed not only text, but the shadows on the snow match the text too. But in this case, it almost forgot about the shadow. So it almost similar to the image with the text cool. It doesn't work perfectly in all possible cases. For example, when the text on the original image is not very recognizable, I tried to replace all the text to nanobanana on all the images. Always keep in mind quality downgrade of the output image. But Nano Banana can be good at removing objects from the scene even in some not so easy cases. The fence from this image was deleted 100% perfectly. But what if there's human face on the image behind the fence? In this case, just small part of the face was covered by the part of the fence. But look at this example. I made few tries with this one. And here's the best result I got. It can't keep the pose 100% the same. When I tried to force Gemini to keep the hand in the same place it removed it with the fence, but the face looks almost the same. When I tried to remove all the people from this image, I got this result from the first try. You can see that there are still some people on the background. So, I edited this picture by asking Nano Banana remove these human figures, too. Here's another image. I wanted to keep the man in the center the same, but remove all the other people from the picture. And again, there were few people on the first generated image, so I edited it with the same prompt. And here's final result. So, as you can see, when there are lots of objects you want to remove at once, you'll probably need to make few steps of image editing. The idea of adding two, three, or even more people on the single image is obvious. Like in this case, for example, I made this guy and a king sit in the kitchen at evening. Faces are very similar to the reference images as well as clothing. But you probably notice that there can be lack of similarity between real face of the person and the generated one. This can happen to just one character on the image, or even both faces can mismatch the references. I found that this annoying error appears much less frequently if you keep the same outfits and other details of the character's appearance as on the reference images. I didn't mention the clothing in my prompts, so the outfits were recreated without any changes. This doesn't mean that you should never change clothing, haircut, makeup, or facial expressions of the characters. But if you add some significant changes, the chances of getting inconsistent faces are higher. If you want more tips on character consistency in Nano Banana, check my previous video about this tool. In that video, I also explained how to generate images faster and more.