How To Master Prompting For Gemini's Google AI Studio Nano Banana

What's going on guys? Michael Elliot here. So today I'm going to run you through master prompting for Nano Banana. I'm going to teach you how to stop LLMs doing the most frustrating thing that they do that is hallucinating. You ask an AI to do something and it does something completely different. Everybody struggled with that issue. I'm going to put that to bed. Today I'm going to show you how you can use it to ensure that nanobanana gives you the correct output, including the correct structures to use. So let's jump in. So let's jump straight into Nano Banana. We head over to aistudio.google.com just log in if you need to. Nano Banana top right. If Nano Banana is not there and you do have something else set like Flash or Pro, just hit Nano Banana, put the temperature down to zero. This is the creative creativity of the AI. Exactly what we're trying to avoid. Hallucination, etc and this gives us the ability now to prompt. And without a prompt this tool is not going to do anything. It's just going to sit looking at us as it is now. So we're going to jump in with a prompt and I've got a couple that are pre created. Now the pre created prompt here is a young man with freckles, smiling thoughtfully, chatting with a colleague, sitting on a sunlit seat in a cozy cafe next to a log fire shot on an A Sony A6, 400. Soft, natural light, warm and inviting seems like a mouthful. But it is the reason why we get images that are the standard of this and this is what I'm going to teach you today. How to be able to do this so we get an outstanding quality image, right? The reason that we got that outstandingly high quality image is, is because we followed this, the prompt followed this exact structure and on the screen I'm going to match it with what the query is that I've just read out to you. So we have the subject, we have the action, we have the environment, we have the style, we have the lighting and we have the details. And every single one of your prompts should follow that. Every exact flow for generating the highest quality standard images possible. By doing that and ensuring that you have all of those boxes ticked, you're going to be able to generate images that look like this on a consistent basis with a consistent quality output instead of it hallucinating and doing something can like ridiculously crazy. So I want to go one step deeper than this to showcase to you how you can have this now I've renamed this already to Nano Banana prompt generator. Okay. And all I want you to do is click this button. It's a bit of an arc at the end. Wasn't on the right line. This button here looks like a notepad. If you hover over it or. Yeah, if you hover over it, it'll show you system instructions, so we can actually give it a brain. Now, if I click in six system instructions, you'll see optimal optional tone and style instructions for the model. Well, I just read out to you the structure that is necessary to get the highest quality image as possible. So in here I'm going to paste a couple of things. So I'm going to say all of your prompts should follow this structure. Subject, environment, style, lighting, details. Okay. I'm then gonna copy in this little prompt that I wrote out, which is how you should speak to LLMs. Always. Nano banana chat gbt5. You're a prompt creator for Nano Banana with years experience in prompt engineering. It's very basic. You could go more in depth, but for the use case of this video, that's good enough. And then what I'm going to do is I'm going to take that example that we've just used and I'm going to pop it right in the middle and then I'm going to bring this up and then I'm going to say, in fact, I'm not. I'm going to pop it there. I'm going to copy it and then put it underneath this. All of your prompts should follow the. For example. And what have we just done? Well, we have. We have given it a. An instruction around who it is. We've told it who it is. You're. You're a prompt creator for Nana Banana, and you have years of experience in prompt engineering. We've then told it that all of its prompts should follow the exact structure that we just said. Then we gave it a direct example of one that we wanted for an image. So now what I'm going to do is I'm going to say, can you please write me a prompt that involves a waterfall? And we're going to fire it off and it's going to give us back a prompt in exactly the same format as what we wrote the other one, which means that the quality of the output from the image will be of similar ilk. Here's your image. A majestic, cascading, crystal clear, surrounded by vibrant. So now, one final time, I'm going to ask it to please create me a prompt for an Image of a hyper realistic waterfall. So I'm helping it a little bit. So we got an exceptional image there that was frighteningly good. But it wasn't hyper realistic, it wasn't real. Now I've said hyper realistic. The prompt should contain something around the, you know, been. Been. Been realistic. Could have gone with photorealistic. But the point is, is that we're using this in such a simple manner but getting a consistent output of prompts which I'm now going to run and this is going to spit out something which is, you know, you can see. Look at the, the quality of that. Like just the prompt in, in general, the water surface should show intricate patterns of foam and spray with subtle refractions of light. The surrounding rocks are wet and glistening. It's taken our instruction and example and beefed it up which is good for the AI model. And we'll see from the output of the, of the image ultimately how good it really is. But you can see by giving the AI a brain and asking it to generate prompts, you can use this in your. Look at that. I mean that is ridiculous. That's like that has been shot with a camera in real life. I mean the rainbow even got a rainbow in it, which is odd. But you have this, what an image that is. Okay. And it's all based off of the fact that we gave it the image prompt in the, in the first place. So we know that we can now prompt engineer Nano Banana to give us the exact output following that exact five step process. And that's going to give us the best quality output. Now we need to hit into editing, editing images that already exist because it is different. And this brings us to our second section which is Edit, update and adding different things to two images. Now I already have an image, I want to upload it. Now. This image is of a man walking in the park that Nano Banana actually created. And I'm going to give it a prompt and I'm going to say please edit this image. Specifically is T shirt. I want it to be green, oversized, yet well fitting jumper. Change the environment to be Nightfall. Let's hit run. You'll notice I've removed the system instructions because this is a completely different beast. You could do it again, but it's a different, different style. Now you can see that he's wearing an oversized green jumper. If we wanted to make further changes to this, it didn't change it to Nightfall. So I'm going to tell it to be. Please ensure that the image is edited so that it's 3am in a dark park. Let's hit run, see what this comes out with. From a perspective of night time. It followed all of my instructions. It updated the image, the image properly. Let's hope this doesn't go too far away from the original. As it goes away and brings it down to night time, you can see it's drawn, it's drawn it, it's drawn it into being dark. It's not quite night time because of the sky. So, so maybe we, we give it a specific example and we. This is, this is a good learning experience because this actual technique follows a different structure. So this technique follows the infographic on the screen now, which is when you are editing an image and not generating a new one. The action words are add, change, remove, make, replace, increase or decrease. Those are the words that we can use to ensure that we get the correct changes. There are many other verbiages that you could probably squeeze in there, but those are the main ones. Add, change, make, remove, replace are the big ones. Okay then when it comes down to the structure, all of the prompts should follow. Action change specific element to edit. So you want to tell it what to edit, the desired style and effect and the relevant details alongside it. So now we're going to go back in here and we're going to ensure that we get the dark night sky that we wanted. So not only am I going to do that, I'm going to show you how you can influence the details to make two changes at once. I wouldn't normally advise it, but it's interesting. I'm showing you the techniques here. Plus, please edit this image specifically is green jumper. I want it to be a white, oversized, well fitting jumper. Also please ensure the sky's edited to be in a full moon. So we're challenging it now because the prompt has changed. The details also can contain an edit, which kind of mixes up our system prompt slightly. But you can see the power of Nano Banana and LLMs in this day and age allows it to be able to do so with great significance. As long as you stick to those system prompting techniques using Nano Banana, you'll get the exact quality images that you want. There's nothing standing in your way from having the highest quality imagery available literally from any business on the planet. The gap has now shrunk. As a small business that didn't that couldn't afford to spend $8,000 on a, on a suite or £8,000 on a suite of images from a creative in central London or Manhattan or wherever like you now have the ability to create them. Follow the system prompting that I've put together in this video for you because that will ultimately determine your success rate. It stops hallucinations, as we've just seen. It does exactly what we want it to do. Based off the back of that, my name is Michael Aliyah. If you've enjoyed this video, please, like, subscribe. Comment. It means more than you know. I'll catch you in the next one.