Nano Banana Prompt to Create Stunning AI Photos

A nano banana prompt is the line of text you paste into Google's Gemini image tool to edit a photo or build one from scratch. People call the model nano banana, the actual name is Gemini 2.5 Flash Image, and lately you will also see folks talk about Nano Banana Pro and even a Nano Banana 2 for the newer, sharper results. The grid at the top of this page is full of prompts you can copy right now. This part is about the words themselves: how a good prompt is put together, why some go viral while yours flops, and how to fix the ones that come out wrong.

What a nano banana prompt actually does

The model reads your photo and your instruction at the same time. So the prompt is half description and half rules. You are telling it what to make, and quietly telling it what to leave alone. That second part is where most people slip. They write a lovely description of a scene and forget to say 'keep my face', so the output looks great and nothing like them.

Plain English works. You do not need code, brackets, or special syntax. There was a whole trend of writing prompts in a JSON style with curly braces and semicolons, and from what people are reporting it barely moves the needle. One person tested it and said the formatting did almost nothing. Write a clear sentence or two instead. The model handles that better than a wall of fake code.

The shape of a nano banana prompt that works

Once you have run a few, you start to see the same skeleton under every good one. I write mine in this order, and it rarely lets me down.

1. The subject. Who or what is in the frame, in three or four words.
2. The keep line. What from your photo must survive, usually the face and skin tone.
3. The look. Film grain, studio light, oil paint, plastic toy, soft editorial color.
4. The setting. Where it happens and how it is lit.
5. The framing. Close portrait, full body, three quarter angle.

Stitch those into a sentence. Something like: a portrait of me, keep my exact face, warm 35mm film look with soft grain, sitting by a window in afternoon light, shot close. That single line covers everything the model needs and nothing it can trip over.

Tip: decide the keep line before the look line. If you nail the likeness first, you can throw any style at it. If you chase the style first, the face usually drifts.

Nano banana prompt styles people are pulling off right now

The looks that get shared most this month lean editorial and filmic rather than the cartoon stuff from a while back. Analog film portraits are everywhere, the kind with grain and a slightly faded color cast. So are soft editorial portraits, ballet and dance themed shots, and clean studio headshots people use for profiles. The toy figurine of yourself, boxed like a collectible, is still going strong because it is easy and it reads instantly.

Here is a quick map you can steal from. Pick a row, copy the wording into Gemini, swap in your own photo.

Keeping yourself recognizable with one nano banana prompt

This is the thing that separates a one-off lucky image from a set you can actually use. People making short videos and photo series with nano banana say the trick is to lock a prompt format and reuse it, changing only the small parts. If you rewrite the whole prompt every time, you are basically rolling dice and the face shifts on every run.

So pick your wording once. Keep the subject and keep line word for word, and only change the scene or the pose. The model holds likeness much better when most of the prompt stays put. When you do need a different angle, add it plainly: same person, same face, three quarter view, looking left. Do not restate the whole style, just nudge the one thing you want different.

How to run a nano banana prompt without fuss

The mechanics are short and they barely change between prompts.

• Open Gemini and upload one clear, well lit photo.
• Paste the prompt from the grid above.
• Add 'keep the same face and skin tone' if a person needs to stay recognizable.
• Generate, then run it two more times and compare.

Results wobble a little run to run, and that is fine, it works for you. The second or third attempt is often the keeper. If you found a winner, save the exact wording somewhere so you can come back to it next week.

When a nano banana prompt fights back

Most failures have a quick fix once you know the pattern.

The face changed too much

Your style words are overpowering the likeness. Soften them, and move 'keep my exact face' to the front of the prompt so the model reads it first.

The skin looks like plastic

Add 'natural skin texture, visible pores' and start from a sharper input. Over-smoothing is almost always the input photo being soft, not the prompt being wrong.

It refused to generate

Nano banana is strict about certain things. Uploaded photos of real people, anything involving celebrities, and images of children get heavy moderation, and it will often just decline rather than try. People comparing it to other tools note that nano banana tends to tell you no upfront instead of generating something and then blocking it. If you hit a wall, change the subject or use a photo you clearly have the right to use.

How a nano banana prompt stacks up against other tools

You can paste similar wording into ChatGPT's image mode, Midjourney, or DALL-E, and a lot of these prompts do carry over. Some of the film portrait prompts going around were actually built in ChatGPT first and then run through Gemini. The difference is mostly in editing an existing photo. Nano banana is unusually good at taking your real picture and keeping the face while changing everything around it, which is why the photo-edit crowd gravitates to it. For pure from-scratch art with no reference, Midjourney still has a distinct look people like. Try the same prompt in two of them and keep whichever gives you the result you want.

One practical note on cost. The prompts here are free to copy, and the Gemini app has a free tier that covers everyday edits. Heavy or batch use runs through the paid API at a few cents an image, so try a prompt free first and only think about paying if you are generating in bulk.

Pick any prompt from the grid, drop in a sharp photo, and run it three times before you judge it. Keep the wording that lands, change one word at a time after that, and you will get a repeatable look instead of a one-time fluke.