Nano Banana 2 and Pro: what an image costs and how to use it

# Nano Banana prompt card · AIGE

How to write prompts for Google's Nano Banana 2, Nano Banana Pro and Nano Lite in AIGE (aige.ws), with the sizes and credit prices AIGE charges.

Updated 2026-09-27 · By AIGE

**How to read the tags.** Every tip comes from Google or fal.ai and is tagged with the model it was written for:
`[NB2]` Nano Banana 2 · `[Pro]` Nano Banana Pro · `[NB2+Pro]` both · `[NB1]` the original Nano Banana (Gemini 2.5 Flash Image), older advice that is still useful. Numbers in brackets point to the Sources list at the end. Anything about menus, sizes and credits describes AIGE itself.

---

## Which one to pick

| | Nano Banana 2 | Nano Banana Pro | Nano Lite |
|---|---|---|---|
| Name in the AIGE menu | Nano 2 | Nano Pro | Nano Lite |
| Google's name | Gemini 3.1 Flash Image [5] | Gemini 3 Pro Image [4] | "Nano Banana Lite" on fal [10] |
| Sizes in AIGE | 0.5K, 1K, 2K, 4K | 1K, 2K, 4K | One size, no size choice |
| Credits per image | 6 · 8 · 16 · 48 | 15 · 20 · 30 | 4 |
| How Google describes it | Its "go-to image generation model" [1] | Made "for professional asset production and complex instructions" [1] | n/a |
| When AIGE suggests it | Everyday images and edits; the default | You want Google's Pro model, or a 4K image | Quick drafts before a final render |
| Reference images, per Google's API docs | Up to 10 object images and up to 4 character images [1] | Up to 6 object, 5 character and 3 style images; 14 in total [1] | n/a |

- At 4K, Pro costs less than Nano Banana 2 in AIGE: 30 credits against 48.
- Results differ by subject, so run the same prompt on both and keep the one you prefer.
- An edit (a run with a photo attached) costs the same as a new image at that size.
- If an image fails, AIGE gives the credits back.

## The prompt formula

Google's core rule: describe the scene in sentences. `[NB2+Pro]` "A simple list of keywords won't cut it; you need to describe the scene narratively" [3]. Google first gave the same advice for the original model `[NB1]` [2].

The five parts Google names `[Pro]` [4] `[NB2+Pro]` [3]:
1. **Subject.** Who or what, described precisely: "a stoic robot barista with glowing blue optics", not "a robot".
2. **Composition.** How the shot is framed: extreme close-up, wide shot, low angle, portrait.
3. **Action.** What is happening in the picture.
4. **Location.** Where it happens, for example "a sun-drenched meadow at golden hour".
5. **Style.** The overall look: photorealistic, watercolor, film noir, 3D animation, 1990s product photography.

Then add detail `[Pro]` [4] `[NB2+Pro]` [3]:
- **Camera and lighting.** "A low-angle shot with a shallow depth of field (f/1.8)", "golden hour backlighting creating long shadows", "three-point softbox setup".
- **Materials.** "Navy blue tweed", not "a suit jacket".
- **Text.** The exact words in quotes and how they should look (see Text inside images).
- **Purpose.** Say what the image is for (a menu, an ad, a product page). Google: "The model's understanding of context will influence the final output" `[NB2]` [1].
- **Open with a verb.** Google suggests starting with a strong verb that names the job: Create, Edit, Combine, Transform `[NB2+Pro]` [3].

**Template** (built from Google's formula [3][4]):

```
[Create / Edit / Combine] a [style] of [specific subject], [action], in [location and time].
[Shot type and camera angle]. [Lighting]. [Lens, film or medium]. [Mood].
[Optional: the text "..." in a [font description], placed [where].]
```

**Keyword list vs scene:**
- Keyword list: `woman, café, laughing, warm, bokeh, 4k`
- Scene: `A photorealistic medium shot of a woman laughing at a small marble table in a sunlit café, holding an espresso cup. Warm late-afternoon light from the window on the left, shallow depth of field so the street outside blurs. Shot on an 85mm portrait lens. Relaxed, candid mood.`

**Aspect ratio and size:** choose them in the AIGE menu, not in the prompt. AIGE sends your chosen ratio with every run.

## Editing a photo

In the AIGE studio, attach the photo, choose Nano 2 or Nano Pro and describe the change. With a photo attached, AIGE sends the run to the model's editing version automatically. Google's editing templates `[NB2]` [1] (first written for `[NB1]` [2]):

| Job | Template |
|---|---|
| Add, remove or change something | `Using the provided image of [subject], please [add/remove/modify] [element] to/from the scene. Ensure the change is [how it should blend in].` |
| Change one part only | `Using the provided image, change only the [element] to [new element]. Keep everything else in the image exactly the same, preserving the original style, lighting, and composition.` |
| New art style | `Transform the provided photograph of [subject] into the artistic style of [art style]. Preserve the original composition but render it with [brushwork, palette, texture].` |
| Combine several images | `Create a new image by combining the elements from the provided images. Take the [element from image 1] and place it with/on the [element from image 2]. The final image should be a [final scene].` |
| Keep a face or product detail exact | `Using the provided images, place [element from image 2] onto [element from image 1]. Ensure that the features of [element from image 1] remain completely unchanged. The added element should [how it fits].` |
| Sketch to finished picture | `Turn this rough [medium] sketch of a [subject] into a [style] photo. Keep the [features] from the sketch but add [materials, colors, details].` |
| Translate text in the image `[Pro]` [4] | `Translate all the English text on the [object] into [language], while keeping everything else the same.` |

Rules that make edits work:
- Say exactly what must stay the same `[NB2+Pro]` [3]. Google's tip for key details: describe them in great detail alongside your edit request `[NB2]` [1].
- When you send several images, give each one a role, for example "Use Image A for the character's pose, Image B for the art style, and Image C for the background environment" `[Pro]` [4].
- Each AIGE run is a new request with no chat memory. Where Google says to refine in follow-up messages [1], do this in AIGE: attach the last result and describe the one next change.
- Set the aspect ratio to match your photo. AIGE sends the ratio from the menu (1:1 unless you change it), so a vertical photo edited at 1:1 comes back square.

## Keeping the same face

**In AIGE.** Save a character from a few photos. In the studio prompt, type `@` and pick the character. Its reference photos are sent with the prompt, ahead of any photos you attach. From Claude or ChatGPT, pass the character's `characterId` (see the last section). Nano 2, Nano Pro and Nano Lite all accept references.

**What Google documents:**
- Nano Banana 2 keeps the resemblance of up to 4 characters, and Pro takes up to 5 character images, in one image `[NB2]` `[Pro]` [1].
- For a character seen from different angles, put earlier results into later prompts. For a complex pose, add a reference image of that pose `[NB2]` [1].
- Say in the prompt that the identity must not change. Google's own example prompts use lines like "Keep the attire and identity consistent of all 3 characters, but their expressions and angles should vary" `[NB2]` [5].
- If features drift after many edits, start again from a detailed description `[NB1]` [2]. In AIGE, start a new run from the saved character instead of editing the latest edit again.
- Google says character consistency "may not always get it right" `[NB2]` `[Pro]` [6][7]. Make a few and keep the closest match.

Tip: because the character's photos go first, refer to your other attachments by what they show ("the photo of the orange sneakers") rather than by number.

## Text inside images

Google's rules `[NB2+Pro]` [3], `[Pro]` [4]:
- **Put the words in quotes:** `the headline "URBAN EXPLORER"`.
- **Describe the font, or name it:** "bold, white, sans-serif", "Century Gothic".
- **Say where it goes:** "rendered ... at the top".
- **Write the text before the image.** Google says the models handle text better if you settle the words first and then ask for an image containing them `[NB2]` [1] [3]. Draft the copy with Claude or ChatGPT, then paste it in quotes.
- **Say which language** the text should be in. Both models can translate and localize text inside an image [3][5].

Honest limits, from Google:
- "Rendering small text, fine details, and producing accurate spellings may not work perfectly" `[Pro]` [4]. Nano Banana 2 "can still struggle with small faces, accurate spelling, and fine details" [7]. Keep key words large (headline size) and read every word before you publish.
- Text in other languages may have grammar mistakes or miss cultural nuances [4][6][7].
- Google's API docs list these languages as the ones the models perform well in: EN, ar-EG, de-DE, es-MX, fr-FR, hi-IN, id-ID, it-IT, ja-JP, ko-KR, pt-BR, ru-RU, ua-UA, vi-VN, zh-CN [1].
- For designs that are mostly text, AIGE's picker also has GPT Image 2 ("GPT2" in the menu). Try the same prompt on both.

## Photoreal vs stylized

**Photoreal.** "Think like a photographer": name the camera angle, lens, lighting and fine details `[NB1]` [2]. Google's newer guide adds `[NB2+Pro]` [3]:
- The camera type changes the feel: a GoPro, a Fujifilm camera, a cheap disposable camera with flash.
- Lens and focus: "wide-angle lens" for scale, "macro lens" for detail, "shallow depth of field (f/1.8)".
- Film and color: "as if on 1980s color film, slightly grainy", "cinematic color grading with muted teal tones".

**Stylized** (illustration, sticker, 3D, anime). Name the style and the medium, and describe the visual details, such as bold outlines, cel-shading and a color palette. Ask for a white background if you need one `[NB2]` [1] `[NB1]` [2].

**From Claude or ChatGPT**, the `style` setting (`photo`, `anime`, `cinematic`, `neutral`) adds a line such as "Visual style: cinematic." to the end of your prompt. You can also describe the style in the prompt itself.

## Aspect ratio and size

| Where | Models | Ratios | Sizes |
|---|---|---|---|
| Studio (prompt bar) | Nano 2, Nano Pro, Nano Lite | 1:1, 4:5, 3:4, 9:16, 16:9, 4:3 | 1K, 2K, 4K (Nano 2 and Pro); Lite has one size |
| Canvas (chat settings) | Nano 2, Nano Pro, Nano Lite | The six above, plus 3:2, 2:3, 21:9 | 0.5K, 1K, 2K, 4K |
| Claude / ChatGPT (MCP) | `nano-banana-2`, `nano-banana-pro`, `nano-banana-lite` | The nine above, plus 9:21 (sent to Nano as 9:16) | `0.5K`, `1K`, `2K`, `4K` |

- 0.5K is for Nano Banana 2 only. A 0.5K request on Nano Pro runs, and is charged, at 1K.
- Nano Lite has no size setting.
- **Plans** (every image is paid in credits): 0.5K and 2K need Starter or higher. 4K needs the Pro plan or higher, because Starter stops at 2K. 1K and Nano Lite are on every plan.

## Common problems and fixes

| Problem | Fix | Source |
|---|---|---|
| Half the prompt is ignored | Write it as a scene, not a keyword list. For busy scenes, go step by step: "First, create a background of... Then, in the foreground, add... Finally, place..." | `[NB2]` [1], `[NB2+Pro]` [3] |
| Something you said "no" to shows up anyway | Describe the result you want instead: "an empty, deserted street with no signs of traffic", not "no cars" | `[NB2]` [1], `[NB2+Pro]` [3] |
| The edit changed too much | Use "change only the..." and list what must stay exactly the same | `[NB2]` [1], `[NB2+Pro]` [3] |
| The face drifts over several edits | Start again from the saved character with a detailed description instead of editing the last edit | `[NB1]` [2] |
| Words are misspelled | Quote the exact words, describe the font, keep text large, check before publishing | `[NB2+Pro]` [3], `[Pro]` [4] |
| You asked for "4 versions" and got one | The prompt cannot set how many images you get. Use the ×1–×4 control in the studio, or `count` from Claude or ChatGPT | `[NB2]` [1], fal [8] |
| The edit comes back square | Set the ratio in the menu to match your photo; AIGE uses 1:1 unless you change it | AIGE |
| The model mixes up your references | Give each image a role ("Image A for the pose, Image B for the style") | `[Pro]` [4] |
| The facts in an infographic are wrong | Check all data yourself. In the studio, the globe toggle on Nano 2 and Nano Pro lets the model search the web for current information (+1 credit) | `[Pro]` [4], `[NB2+Pro]` [3], fal [8][9] |
| Exploring ideas uses up credits | Draft on Nano Lite (4 cr) or Nano 2 at 0.5K (6 cr, on the Canvas or from Claude/ChatGPT), then make the final at 1K–4K | AIGE |

Note on web search: Google says Nano Banana 2's search grounding "does not support using real-world images of people from web search at this time" `[NB2]` [1].

## 10 example prompts

Credits are per image. "cr" means credits.

**1. Product on a plinth.** Nano 2 · 2K · 4:5 · 16 cr
> A high-resolution, studio-lit product photograph of a frosted glass perfume bottle with a brushed gold cap, standing on a white travertine plinth against a warm sand-colored backdrop. The lighting is a three-point softbox setup that creates soft highlights along the glass and a gentle shadow to the right. The camera is at eye level, slightly low, to show the full silhouette. Ultra-realistic, with sharp focus on the cap's brushed texture.

**2. Lifestyle portrait of your saved character.** Nano 2 · 1K · 4:5 · 8 cr. Type `@` and pick the character; it shows here as @Mia.
> @Mia A photorealistic medium shot of the woman from the reference photos sitting at a small marble table in a sunlit Lisbon café, laughing mid-conversation with an espresso cup in her hand. Warm late-afternoon window light from the left, shallow depth of field so the street outside blurs. Shot on an 85mm portrait lens. Keep her face and features exactly as in the reference photos. Relaxed, candid mood.

**3. Flat-lay.** Nano 2 · 1K · 1:1 · 8 cr
> A top-down flat-lay photograph of a weekend travel kit on a pale linen sheet: a tan leather passport holder, folded sunglasses, a small film camera, a linen shirt folded into a neat square and a sprig of dried lavender. The items are evenly spaced with generous empty space around them. Soft, even daylight from above that keeps shadows minimal. Calm, editorial mood.

**4. Poster with a headline.** Nano Pro · 2K · 3:4 · 20 cr
> Create a poster for an outdoor summer jazz night. The headline "JAZZ ON THE ROOF" is rendered in large, bold, cream-colored condensed sans-serif letters across the top third. Below it, a saxophonist silhouetted against a deep orange sunset over city rooftops, in a flat screen-print style with visible grain. At the bottom, the line "Friday 8 PM" in a smaller, clean sans-serif font. Limited palette of orange, navy and cream. No other text.

**5. Outfit swap (edit).** Nano 2 · 1K · 3:4 · 8 cr. Attach a photo of the person and a photo of the coat.
> Create a professional e-commerce fashion photo. Take the green wool coat from the photo of the coat and let the woman from the photo of the woman wear it. Generate a realistic full-body shot of her wearing the coat, with the lighting and shadows adjusted to match her original street setting. Keep her face, hair and pose unchanged.

**6. Background replacement (edit).** Nano 2 · 1K · the photo's own ratio · 8 cr. Attach the photo.
> Using the provided photo, change only the background to a bright, minimalist studio with a white wall and a soft shadow on the floor. Keep the person, their clothing and their pose exactly the same, preserving the original lighting on the person.

**7. Sticker or illustration.** Nano Lite · 1:1 · 4 cr for drafts; run your favorite again on Nano 2 for the final
> A kawaii-style sticker of a sleepy corgi curled up inside a coffee cup. The design features bold, clean outlines, simple cel-shading and a soft pastel color palette. The background must be white.

**8. Food menu shot.** Nano 2 · 2K · 4:3 · 16 cr
> A high-resolution food photograph of a bowl of shoyu ramen for a restaurant menu: glossy broth, a halved soft-boiled egg with a jammy yolk, two slices of chashu pork, nori and sliced scallions in a dark ceramic bowl on a walnut counter. Shot from a 45-degree angle with a 50mm lens. A warm key light from the back left makes the steam visible, and the background falls into soft shadow. Appetizing, cozy mood.

**9. Interior scene for print.** Nano Pro · 4K · 16:9 · 30 cr (Pro plan or higher)
> A photorealistic wide-angle shot of a Scandinavian living room at blue hour: a low oatmeal bouclé sofa, a round oak coffee table with a stack of art books, a paper floor lamp giving warm light, and floor-to-ceiling windows showing a snowy pine forest outside. Eye-level camera from the corner of the room, 24mm lens, everything in focus. Quiet, cozy mood.

**10. Multi-reference composite.** Nano Pro · 2K · 16:9 · 20 cr. Pick @Mia, then attach a photo of the sneakers and a photo of the skate park.
> @Mia Combine the provided images into one photo. Use Mia's reference photos for her face and hair, the photo of the orange sneakers for the shoes she wears, and the photo of the concrete skate park for the location. Show her mid-stride walking toward the camera at golden hour: full-body shot, low angle, with the light and shadows on her and the sneakers matching the late sun in the skate park. Keep her face and the sneaker design unchanged.

## Settings and prices in AIGE

| Model | Size | Credits | Credit key charged | Where | Plan |
|---|---|---|---|---|---|
| Nano 2 | 0.5K | 6 | `generateCharacter05k` | Canvas, MCP | Starter+ |
| Nano 2 | 1K | 8 | `generateCharacter1k` | Studio, Canvas, MCP | All |
| Nano 2 | 2K | 16 | `generateCharacter2k` | Studio, Canvas, MCP | Starter+ |
| Nano 2 | 4K | 48 | `generateCharacter4k` | Studio, Canvas, MCP | Pro+ |
| Nano Pro | 1K | 15 | `generateCharacterPro1k` (studio edit: `refineCharacterPro1k`) | Studio, Canvas, MCP | All |
| Nano Pro | 2K | 20 | `generateCharacterPro2k` (studio edit: `refineCharacterPro2k`) | Studio, Canvas, MCP | Starter+ |
| Nano Pro | 4K | 30 | `generateCharacterPro4k` (studio edit: `refineCharacterPro4k`) | Studio, Canvas, MCP | Pro+ |
| Nano Lite | one size | 4 | `generateCharacterLite` | Studio, Canvas, MCP | All |
| Web search toggle (globe) | add-on | +1 | `webSearchSurcharge` | Studio, Nano 2 and Nano Pro | Starter+ |

- Making ×2, ×3 or ×4 images charges the price per image that many times.
- On the Canvas, images are made by the Canvas assistant. Its message is billed separately, from 2 credits (`agentTurn`).
- The studio shows the credit cost in its settings panel before you run.

## Use it from Claude or ChatGPT

AIGE runs an MCP server at `https://aige.ws/mcp`. Setup steps are at aige.ws/mcp-info, and API keys are at aige.ws/api-keys. The image tool is `antix_generate_image`:

| Parameter | Values |
|---|---|
| `prompt` | Your prompt, up to 2,000 characters (required) |
| `model` | `nano-banana-2` (default), `nano-banana-pro`, `nano-banana-lite` (other image models are also available) |
| `resolution` | `0.5K`, `1K`, `2K`, `4K` |
| `aspectRatio` | `1:1`, `16:9`, `9:16`, `4:5`, `3:4`, `4:3`, `3:2`, `2:3`, `21:9`, `9:21` (default 1:1) |
| `count` | 1–4 |
| `characterId` | A saved character from `antix_list_characters` |
| `characterUse` | `face` (default), `body` or `both` |
| `referenceImages` | Up to 10 image URLs or data URLs (`referenceImage` for a single one) |
| `style` | `photo`, `anime`, `cinematic`, `neutral` |

Generation runs in the background. The tool returns a `taskId`, and you fetch the result with `antix_check_task`. With a character, its photos go first, then your references, 10 at most in total.

**Example ask:**
> Using AIGE, make a 4:5 photo of my character Mia on a rooftop café at golden hour. Use Nano Banana 2 at 2K. Write the prompt as a full scene with lens and lighting, and keep her face as in her saved photos.

The call Claude makes (after it looks up Mia's id with `antix_list_characters`):

```json
{
  "model": "nano-banana-2",
  "prompt": "A photorealistic medium shot of the woman from the reference photos sitting at a rooftop café table at golden hour, looking out over the city with a relaxed smile. Warm low sun from behind her creates a soft rim light on her hair; shallow depth of field blurs the skyline. Shot on an 85mm portrait lens. Keep her face and features exactly as in the reference photos.",
  "characterId": "<Mia's id>",
  "aspectRatio": "4:5",
  "resolution": "2K",
  "count": 1
}
```

Cost: 16 credits (`generateCharacter2k`). If the image fails, the credits are refunded automatically.

## Sources

1. Google AI for Developers: Nano Banana image generation (Gemini API docs; the example prompts run on Nano Banana 2; last updated 2026-09-23). https://ai.google.dev/gemini-api/docs/image-generation
2. Google Developers Blog: prompting guide for Gemini 2.5 Flash Image (Aug 28, 2025; written for the original Nano Banana). https://developers.googleblog.com/en/how-to-prompt-gemini-2-5-flash-image-generation-for-the-best-results/
3. Google Cloud Blog: The ultimate Nano Banana prompting guide (March 2026; written for Nano Banana 2 and Nano Banana Pro). https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-nano-banana
4. Google Blog: 7 tips to get the most out of Nano Banana Pro (Nov 20, 2025). https://blog.google/products-and-platforms/products/gemini/prompting-tips-nano-banana-pro/
5. Google Blog: Nano Banana 2: Combining Pro capabilities with lightning-fast speed (Feb 26, 2026). https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/
6. Google DeepMind: Gemini 3 Pro Image (Nano Banana Pro). https://deepmind.google/models/gemini-image/pro/
7. Google DeepMind: Gemini 3.1 Flash Image (Nano Banana 2). https://deepmind.google/models/gemini-image/flash/
8. fal.ai: Nano Banana 2 API (`fal-ai/nano-banana-2`, `/edit`). https://fal.ai/models/fal-ai/nano-banana-2
9. fal.ai: Gemini 3 Pro Image Preview API (`fal-ai/gemini-3-pro-image-preview`, `/edit`). https://fal.ai/models/fal-ai/gemini-3-pro-image-preview
10. fal.ai: Nano Banana Lite API (`fal-ai/nano-banana-lite`, `/edit`). https://fal.ai/models/fal-ai/nano-banana-lite
11. AIGE MCP setup. https://aige.ws/mcp-info