# nano banana photo: how to make and edit a picture

> A nano banana photo is an edit or a new picture from Google's Nano Banana models, driven by a text brief: upload a shot and write what to change and what must stay. Below are five workflows with prompts and before/after results: a portrait from a selfie, a new background, removing extras, restoring an old print and a room.

Published 2026-09-30 · Updated 2026-09-30
URL: https://ai2face.com/blog/nano-banana-foto

Nano Banana — Google's models for generating and editing images — treats a photo like an editor you brief in words. You don't paint a mask or hunt for filters: you upload a shot and write "replace the background with a studio one, leave the person alone". Below are the five tasks people look up most, with prompts and results. Every example was made in [Nano Banana 2](/models/nano-banana-2); the same moves work across the [Nano Banana family](/models/nano-banana).

## What you can do with a photo

Google describes editing as a dialogue: the model adds, removes and changes elements, carries style and color across, and tries to keep the rest of the frame. In practice, photo tasks fall into seven types.

| Task | What to write in the prompt | Which version fits |
|---|---|---|
| Portrait from your photo | who to keep + a new scene, clothes, light | Nano Banana 2, Pro |
| Background swap | "change only the background to …, don't change the person" | any |
| Removing extras | who or what to remove, and what should fill the gap | Nano Banana 2 |
| Changing clothes or an object | "replace only the jacket with …" | Nano Banana 2 |
| Enhancement and restoration | what to remove (scratches, noise), what to keep | Nano Banana 2, Pro |
| Interior from a room photo | one change per step: walls, then furniture | Nano Banana 2 |
| Style transfer | "turn the photo into watercolor, keep the composition" | any |

The rule for every row: name what must stay untouched first, then what must change. Without that, the model may "improve" the face, the pose or the color where you never asked.

## How to make a nano banana photo from your own shot

A portrait in a new scene is the most common request. A regular daylight selfie works, as long as the face is fully visible and not covered by glasses or hair.

1. Upload the photo as a reference.
2. Lock identity in the first sentence: face, features, hair.
3. Describe the new frame: place, clothes, light and lens.
4. If something is off, ask to fix one detail in the result you got.

```
Use the woman from this photo. Keep her face, freckles, hair and features exactly the same.
Make a professional portrait: she stands on a rooftop at golden hour, in a cream linen
blazer over a white top, the city blurred behind her, soft warm backlight,
85mm lens, f/1.8, natural skin texture, a calm confident smile
```

![Left: a woman's selfie in a home kitchen; right: the same woman in a cream blazer on a rooftop at golden hour](https://ai2face.com/_astro/selfie-portrait.DfJE7BSQ_7YkvB.jpg "A kitchen selfie and a portrait from it, in one prompt. Freckles, eye color and hair stayed. Nano Banana 2")

To get a dozen frames with one face, don't upload the selfie every time. AI2Face has a character builder for that: upload 1–3 photos once, then pick a location, clothes and a pose from ready-made assets.

## How to replace the background in a photo

For a background swap, the Gemini API docs include a "semantic mask" template: you don't draw a mask, you name the region in words. You can write the prompt in your language. It goes like this:

```
In this photo, change only [what] to [what]. Leave everything else exactly the same:
the style, the light and the composition.
```

If the background is cluttered and gets in the way, ask to replace the whole surroundings, and list what around the person goes too: the desk, the cables, the papers. Otherwise the model leaves some objects in the foreground.

```
Replace the entire office around him, including the desk, cables and papers, with a seamless light-grey
studio background and soft even light. Make a waist-up business portrait. Don't change the man, his face,
beard, hair, dark navy sweater or expression
```

![Left: a man at a cluttered office desk under fluorescent lights; right: the same man on a grey studio background](https://ai2face.com/_astro/background.4JfVgLIP_YqinI.jpg "A desk photo → a business portrait for a profile. Same face and sweater, studio light. Nano Banana 2")

The same move works for marketplace cards with the wording "clean white background, a soft shadow under the object". For portraits with loose hair, add "keep the hair outline" — the edges come out more natural.

## How to remove extras or change the clothes in a photo

Removing an object is the same semantic mask: name what to remove and hint at what should fill the space. The more precise the description ("the woman in the yellow jacket on the right", not "people"), the less the model touches the rest.

```
Remove only the woman in the yellow jacket and the green bin. Fill the space with the promenade, the railing
and the sea, continuing naturally behind them. Leave the man, the people in the distance, the light and
everything else exactly the same
```

![Left: a man on a promenade, with a bin and a passer-by in a yellow jacket; right: the same frame without the bin or the passer-by](https://ai2face.com/_astro/remove-objects.JUeyiC3K_Z1bcjXu.jpg "The passer-by and the bin removed in one edit. The man, the sunset and the people in the background stay. Nano Banana 2")

Clothes change the same way: "replace only the white shirt with a dark navy denim jacket, don't change the pose or the face". If you need a specific item, upload its photo as a second reference and label the roles: "the first photo is the person, the second is the jacket".

## How to improve and restore an old photo

Restoration is a strength of Nano Banana. The model removes creases, scratches and stains, rebuilds torn edges, and colorizes the print if you ask. List the defects, and separately forbid changes to the faces.

```
Restore this damaged old photo: remove the creases, scratches, dust and the water stain,
rebuild the torn corner. Colorize it with natural, realistic colors. Leave the faces, expressions,
clothing, house and composition exactly the same. Output only the photograph itself, filling
the frame, with no paper border
```

![Left: a faded black-and-white photo of an elderly couple on a porch, with creases and a torn corner; right: the same photo restored and in color](https://ai2face.com/_astro/restore.WTZazKgE_Z1ldmbE.jpg "A 1960s scan with creases and a torn corner → a restored color print. Nano Banana 2")

Check the result in the details. The model fills in what the original doesn't show — the color of the clothes, the pattern on a sweater, the background behind a torn corner. That is acceptable for a family archive, and it is not acceptable for a document.

## How to restyle an interior from a room photo

Change an interior in steps, as in Google's example: walls first, then furniture, then decor. Each step is a separate request sent to the latest result. One edit at a time gives the model fewer reasons to rebuild the whole room.

```
Step 1. Paint only the wall with the window a deep sage green. Leave the floor, the window, the side
walls, the light and the camera angle exactly the same.

Step 2. Furnish the room: a light-grey fabric sofa against the left wall, a round oak
coffee table, a cream wool rug and a tall ficus in a ceramic pot by the window.
Leave the green wall, the window, the floor, the light and the camera angle exactly the same.
```

![Left: an empty room with white walls, and the same room with a sage wall; right: the same room with a sofa, a round table, a rug and a ficus](https://ai2face.com/_astro/interior-steps.6kt7qkxE_Z1CQ6jm.jpg "Empty room → wall painted → furniture placed. The window, floor and camera angle did not shift across two edits. Nano Banana 2")

This is a practical way to try a wall color before a renovation, or to show a buyer how an empty apartment could look. The model does not respect furniture dimensions, so a measured floor plan is what you need for a layout.

## Which Nano Banana version is better for photos

All four models in the family edit photos, but they handle difficult tasks differently.

| Model | Official name | Resolution | Best for |
|---|---|---|---|
| [Nano Banana](/models/nano-banana) | Gemini 2.5 Flash Image | 1K | quick edits: background, style, small details |
| [Nano Banana 2](/models/nano-banana-2) | Gemini 3.1 Flash Image | up to 4K | the default choice: portraits, restoration, interiors, chains of edits |
| [Nano Banana 2 Lite](/models/nano-banana-2-lite) | Gemini 3.1 Flash-Lite Image | 1K | quick single edits |
| [Nano Banana Pro](/models/nano-banana-pro) | Gemini 3 Pro Image | up to 4K | complex scenes, several people, text on a photo |

Gemini's documentation specifically warns that Nano Banana 2 Lite is not built for several references or multi-step edits. For a portrait from a selfie, or an interior in three steps, use Nano Banana 2. Save Pro for frames with several people, or where lettering matters: the model holds a likeness for up to five people.

## Limits and common mistakes

- **You didn't say what to keep.** The model redraws the whole frame and changes the face, the pose or the color. The fix is a first line: "… leave it exactly the same".
- **Several edits in one request.** "Change the background, the clothes and the light, and add a dog" gets done selectively. Do one or two edits per step.
- **A vague description.** "Remove the people" can remove the wrong ones. Name the object by color and by its place in the frame.
- **Recognizable people and brands.** Google's models reject requests with celebrities and logos. In the AI2Face generator you can rerun that generation on [Seedream](/models/seedream).
- **Age.** In the Gemini app, image editing is not available to users under 18.
- **Labeling.** Every Nano Banana image carries an invisible SynthID watermark, and the Gemini app adds a visible logo as well.

You can write the prompt in your language. Put any lettering that should appear on the picture in the language it should be read in, and in quotes. How to structure a request is covered in [Nano Banana prompts](/blog/nano-banana-pro-prompts).

## Conclusions

Nano Banana replaces Photoshop on tasks where an edit is easier to describe than to paint: a new background, removing passers-by, restoration, a portrait from a selfie, trying an interior. The working formula is the same for every scenario — first what to keep, then what to change, and no more than one or two edits per step. The main model for photos is Nano Banana 2; Pro is for scenes with several people and with text. Precise pixel-level skin retouching and RAW work still belong in a classic editor. If official Gemini is unavailable, read about free access in the [Nano Banana free](/blog/nano-banana-besplatno) guide, and try the prompts from this article in the AI2Face generator.

## FAQ

### How do you make a nano banana photo?

Upload a shot as a reference and describe the result: who to keep, what to change, and the light and framing you want. For a picture from scratch, text is enough: subject, scene, light and lens. If the result is almost there, edit it with the next request instead of rewriting the prompt.

### Can you improve photo quality with Nano Banana?

Yes, but the model does not restore the original pixels — it redraws the frame. It removes noise, scratches and creases, evens the light and returns the result at the resolution you chose, up to 4K on Nano Banana 2 and Pro. After that edit, check small details such as lettering and patterns.

### Does Nano Banana change the face while editing?

Sometimes, if the prompt never says to keep the face. Write it explicitly: «don't change the face, hair or age». For a series of frames with one person, the AI2Face character builder is easier: the face stays without uploading the photo again.

### Is there a free Nano Banana photo editor?

Yes. Edits in the Gemini app are available without a subscription. Limits and ways in are covered in the Nano Banana free guide. In AI2Face the models are in the generator: new accounts get welcome credits, and the cost is shown before you generate.

### Can you animate a photo in Nano Banana?

No. Nano Banana makes images only. To turn a photo into video you need a video model: in AI2Face you can set the edited frame as the first frame of a clip.

## Sources

- https://ai.google.dev/gemini-api/docs/interactions/image-generation
- https://blog.google/products-and-platforms/products/gemini/nano-banana-tips/
- https://blog.google/products-and-platforms/products/gemini/updated-image-editing-model/
- https://support.google.com/gemini/answer/14286560