# Nano Banana 2 Image Generator

> Nano Banana 2 is Google's image model.

Canonical: https://www.makeviral.ai/tools/nano-banana-2-image-generator
Updated: 2026-09-17
Published: 2026-09-17
Author: MakeViral Team (https://www.makeviral.ai/about)
Publisher: MakeViral (https://www.makeviral.ai)

Nano Banana 2 is Google's image model. It makes images from a text prompt or edits an existing one, at 1K, 2K or 4K, and it accepts up to 14 reference images on the edit path. On MakeViral it costs 2 credits at 1K and usually returns in 10 to 30 seconds.

Google's image model, in a browser tab, with the reference-image dropzone that makes it useful: up to 14 photos in, one new image out.

**Start now:** [Generate with Nano Banana 2](https://www.makeviral.ai/photo?model=nano-banana-2) (opens the MakeViral app). Part of the [AI Photo Generator](https://www.makeviral.ai/ai-photo-generator).

## What is Nano Banana 2?

It is Google's image generation model, and MakeViral runs it as one of three options inside the [AI photo generator](https://www.makeviral.ai/ai-photo-generator). Nothing to install, no API key. Open the [photo tool](https://www.makeviral.ai/photo?model=nano-banana-2), pick the model, write a prompt, press generate.

The specifics, so you can plan around them.

| Capability | Nano Banana 2 |
|---|---|
| Modes | Text to image, and image to image |
| Resolutions | 1K, 2K, 4K |
| Aspect ratios | 9:16, 16:9, 1:1, 4:3, 3:4, 4:5, 3:2, 2:3 |
| Prompt ceiling | 4,000 characters |
| Reference images | Up to 14 on the edit path |
| Typical generation | 10 to 30 seconds |
| Credits | 2 at 1K, 3 at 2K, 4 at 4K |
| Images per run | 1 to 4 |

Eight aspect ratios is more than the other two offer, and 4:5 matters most: it is the tallest crop the Instagram and Facebook feed shows without cropping you.

The prompt box counts characters against the 4,000 ceiling, and the chips put 9:16 first. Your images are listed newest first with Download, Use as reference, Remix and Delete.

## What is Nano Banana 2 actually good at?

Three jobs, in order of how far they separate it from the alternatives.

**People.** Faces, hands and skin are where image models give themselves away, and this is the one to reach for when a person is in the frame. It is the default behind the persona portraits on the [AI influencer generator](https://www.makeviral.ai/ai-influencer-generator), where a creator portrait costs 2 credits.

**Product shots.** It holds material and finish well: brushed metal, frosted glass, matte cardboard, worn leather. With reference images of your real packaging it is the practical choice for the whole [product photo](https://www.makeviral.ai/tools/ai-product-photo-generator) and [UGC photo](https://www.makeviral.ai/tools/ugc-photo-generator) workflow.

**Edits with references.** The real reason to pick it. Fourteen slots takes a product, a person, a room and a style board in one prompt, or one photo and a single change.

Where it is not the answer: legible text inside the image. If your poster or price tag has to be readable word for word, [GPT Image 2.5](https://www.makeviral.ai/tools/gpt-image-2-5-generator) is the better bet. And if you only need a background, paying 2 credits for what [Z-Image](https://www.makeviral.ai/tools/z-image-generator) does for 1 is waste.

## How do up to 14 reference images work?

Drop images into the reference zone and the run switches to the edit path. The prompt then describes what to do with them, not what to invent. Three recipes cover most of it.

**One reference, one change.** Say what changes and, more importantly, what stays. "Keep the face, hair, pose and background exactly as they are" stops an edit becoming a new image.

**Two to four references, one composite.** Number them: reference 1 is the product, reference 2 the person, reference 3 the room. Then say where each goes and which sets the light.

**Six to fourteen references, one consistent look.** The front, back and cap of your packaging, two shots of your model, a few style frames. This is how a set looks like one campaign.

| References | Typical use | What to write |
|---|---|---|
| 1 | Relight, outfit swap, object removal | What to preserve, then the change |
| 2 to 3 | Product in hand, product in a room | Which reference is which, whose light wins |
| 4 to 8 | Multi-angle packaging, one persona | Which attributes carry across images |
| 9 to 14 | A campaign style board | The look to hold, plus the new scene |

Any generated image can become a reference with Use as reference, so you can chain: the persona, then the product shot, then the slide.

## Which resolution should you pick, and what does it cost?

A credit is about $0.071 on the $49.99 plan for 700 credits, and about $0.067 on the $199.99 plan for 3,000 credits.

| Resolution | Credits per image | Four images | On the $49.99 plan | On the $199.99 plan |
|---|---|---|---|---|
| 1K | 2 | 8 | about $0.57 per run | about $0.53 per run |
| 2K | 3 | 12 | about $0.86 per run | about $0.80 per run |
| 4K | 4 | 16 | about $1.14 per run | about $1.07 per run |

Pick 1K for a feed post, a story, a thumbnail or a slideshow slide. At those sizes the extra pixels do nothing and you pay 50% more for them.

Pick 2K when the image gets cropped in hard, sits behind text, or is printed small. The extra credit is insurance against a soft crop.

Pick 4K for a banner, a print piece, or a shot where texture is the subject. It doubles the run cost, so use it on the winner, not on the search.

The efficient path: draft on Z-Image at 1 credit, find the composition, then re-run the winning prompt here at 1K with references attached. Plans are on the [pricing page](https://www.makeviral.ai/pricing).

## 8 Nano Banana 2 prompts you can paste today

Written for this model at 1K unless noted. The ones that say "using the attached" belong on the edit path.

**1. A creator portrait**
"A 24-year-old woman with freckles and shoulder-length curly auburn hair in a cream ribbed sweater against a sunlit apartment wall, soft window light from the left, eye level on a 50mm lens, shallow focus, calm half smile, 9:16."
Why: age, hair, garment and lens give you a person you can reproduce later.

**2. The same person, a second scene**
"Using the attached portrait as the reference for the face, hair and sweater: the same woman at an outdoor cafe table with a coffee cup, late afternoon side light, other tables blurred behind, same expression, 9:16."
Why: naming the attributes to carry over keeps a persona consistent.

**3. Product in hand, from two references**
"Using reference 1 for the product and reference 2 for the person: she holds the tube at chest height facing camera, label toward the lens and readable, warm kitchen light, blurred background, natural skin texture, 9:16."
Why: numbering the references tells the model which one is the product.

**4. Outfit swap on a photo you own**
"Using the attached photo: replace the jacket with an oversized olive canvas chore coat with visible stitching and two chest pockets, keep the face, hair, pose and background exactly as they are, same lighting direction, 9:16."
Why: the preserve clause is what separates an edit from a new image.

**5. Relight a shot you already have**
"Using the attached product photo: keep the bottle, label and proportions identical, replace the lighting with hard low afternoon sun from the right, a long shadow to the left across a concrete ledge, cool shadow tones, 9:16."
Why: relighting a photo you own beats describing the product again.

**6. Composite a product into a room**
"Using reference 1 for the product and reference 2 for the room: place the diffuser on the wooden shelf at the left third, match the room's soft morning light and its shadow direction, add a faint contact shadow under the base, 4:5."
Why: matching the shadow direction makes a composite read as one photograph.

**7. Two people, candid**
"Two friends in their late twenties on an apartment balcony at golden hour, one leaning on the railing holding a can, the other laughing mid-sentence, warm backlight with a soft flare from the right, framed left of center, 9:16."
Why: a separate action each prevents two identical figures standing still.

**8. Texture at 4K**
"A ceramic pour-over dripper on a raw concrete plinth in an empty gallery, one hard overhead skylight beam, deep falloff into shadow, fine dust visible in the light, centered with large negative space above, 4:3, at 4K."
Why: 4K costs 4 credits, so spend it where texture is the subject.

## Nano Banana 2, GPT Image 2.5 or Z-Image?

Read the second column first. How many of your own images can go into the prompt is the axis these three really differ on.

| Model | Reference images in | What that buys you | Resolutions | Credits at 1K |
|---|---|---|---|---|
| Nano Banana 2 | Up to 14 | Composites, persona consistency, edits that preserve a photo | 1K, 2K, 4K | 2 |
| GPT Image 2.5 | 1 | A single-source edit, and the best legible text of the three | 1K, 2K, 4K | 2 |
| Z-Image | None | Nothing from your world, text prompt only | 1K only | 1 |

**The verdict: pick Nano Banana 2 whenever the image must contain something that already exists.** Your product, your model, your room, your last frame. One slot handles an edit, fourteen handles a campaign.

Switch to GPT Image 2.5 when words inside the image must be readable, and to Z-Image when the image is scenery nobody will inspect.

At 1K both Nano Banana 2 and GPT Image 2.5 cost 2 credits, so that choice is never about money. It is references versus text rendering.

## What Nano Banana 2 will not do here

The limits are the same across all three models here. Know them before you build a workflow on them.

- **No background removal on an uploaded photo.** No cutout tool, no transparent PNG export. You can describe a plain white studio scene and generate into it, which is a new image rather than your photo stripped out.
- **No brand font control.** You cannot upload a typeface. Lettering is the model's guess at a shape.
- **No exact color matching.** No hex field. Generated color drifts between runs, so treat it as a mood.
- **No guarantee a logo or a label renders correctly.** Check every time. Read every word at full size and regenerate when it is wrong. Three attempts is about 43 cents on the $49.99 plan.

MakeViral also does not upload, schedule or post anything, does not connect social accounts and offers no API.

Where the images go next: the [AI slideshow generator](https://www.makeviral.ai/ai-slideshow-generator) builds TikTok photo-mode posts and carousels, and the [AI influencer generator](https://www.makeviral.ai/ai-influencer-generator) turns a portrait into a product-in-hand still. The [Shopify product photo guide](https://www.makeviral.ai/blog/ai-product-photos-for-shopify) covers listing images, and the [TikTok guide](https://www.makeviral.ai/blog/how-to-go-viral-on-tiktok) covers what to do with them.

## Nano Banana 2 Image Generator examples

Real outputs from the app, exactly as they download.

- **Coffee mug on marble** (Z-Image, 1K, 9:16, text to image). Brief: "A matte black ceramic coffee mug on a white marble countertop, soft morning window light, fresh steam, minimal Scandinavian kitchen.". File: https://makeviralai.s3.us-east-1.amazonaws.com/examples/photo/photo-coffee-mug.png. 1 credits. Made 2026-09-17.
- **Serum bottle, readable label** (GPT Image 2.5, 1K, 9:16, text to image). Brief: "A matte black skincare bottle on a white marble slab, soft studio light from the left, water droplets on the glass, sharp label.". File: https://makeviralai.s3.us-east-1.amazonaws.com/examples/photo/photo-serum-label.png. 2 credits. Made 2026-09-18.
- **Cold brew bottle** (GPT Image 2.5, 1K, 9:16, text to image). Brief: "An amber glass bottle of cold brew on a light oak table, soft window light, minimal label that reads COLD BREW, 250 ml.". File: https://makeviralai.s3.us-east-1.amazonaws.com/examples/photo/photo-cold-brew.png. 2 credits. Made 2026-09-18.
- **Cafe photo, relit** (Nano Banana 2, 1K, 9:16, one reference photo). Brief: "Keep the same scene and subject, relight it as a warm golden hour photo and add a small potted plant on the right side.". File: https://makeviralai.s3.us-east-1.amazonaws.com/examples/photo/photo-cafe-relight.png. 2 credits. Made 2026-09-17.

## What the nano banana 2 image generator does for you

- **Up to 14 reference images.** Hand it a product, a person, a room and a style board, then ask for the composite.
- **Built for people.** Faces, hands and skin hold up, which is why persona portraits default here.
- **Eight aspect ratios.** 9:16, 16:9, 1:1, 4:3, 3:4, 4:5, 3:2 and 2:3, with 9:16 first.
- **4,000-character prompts.** Room for product, surface, light, camera and frame, with a live counter.
- **10 to 30 seconds.** Fast enough to iterate, at 2 credits an image at 1K and four images a run.

## How it works

1. **Open the photo tool on this model.** Pick the aspect chip and the resolution. 9:16 and 1K suit short form.
2. **Write the prompt, attach references.** Drop up to 14 photos in the reference zone and number them so each has a job.
3. **Generate, check, download.** One to four images a run, newest first, with Download, Remix and Delete.

## Who it is for

Product photography · AI creator portraits · UGC ad creative · Thumbnails and covers · Ecommerce listings · Slideshow hero frames · Fashion and apparel · Agencies

## Frequently asked questions

### What is Nano Banana 2?

Google's image generation model. It makes an image from a text prompt, or edits images you upload. It runs in the browser here with no install and no API key, at 1K, 2K or 4K, across eight aspect ratios, one to four images per run.

### How many reference images can I use?

Up to 14 on the image to image path. Drop them in the reference zone and number them in the prompt, so the model knows which is the product, which is the person and which sets the light. GPT Image 2.5 accepts one.

### How much does a Nano Banana 2 image cost?

2 credits at 1K, 3 at 2K and 4 at 4K. A credit is about $0.071 on the $49.99 plan for 700 credits and about $0.067 on the $199.99 plan for 3,000 credits, so a four-image run at 1K is about 57 cents.

### How long is the prompt allowed to be?

4,000 characters, with a live counter in the prompt box. That is enough for the product, the surface, the light, the camera and the framing in one prompt, which is how a usable prompt is built.

### How long does one image take?

Usually 10 to 30 seconds, which is fast enough to iterate on wording rather than commit to your first attempt. Generated images appear newest first with Download, Use as reference, Remix and Delete.

### Is Nano Banana 2 good at text inside an image?

It is the weaker choice for that. It approximates lettering rather than typesetting it, so words come back misspelled. If a label, poster or price tag has to be readable, use GPT Image 2.5 and still check every word.

### Can it keep the same character across several images?

Largely, if you work through references. Generate the portrait, then use it as a reference for each new scene and name the attributes to carry over, such as face, hair and garment. Expect small drift, and regenerate frames that wander.

### Can I edit a photo I already own?

Yes. Upload it, then write what changes and what stays. A clause such as keep the face, hair, pose and background exactly as they are keeps an edit from becoming an unrelated new image. There is no background removal tool.

## Viral AI video templates

- [Hotel Lobby AI: Make the Hotel Lobby Trend Video](https://www.makeviral.ai/templates/ai-hotel-lobby-trend): You and a friend, or your two pets, trade a rap into one hanging mic in a bright orange booth. Step into the real clip or the viral cat version with the original song, or get a new 10-second version with an original beat.
- [AI Dog Dancing Video: Make Your Pet Dance](https://www.makeviral.ai/templates/ai-dancing-dog-video): Your dog or cat stands up on its hind legs at home and does a happy, bouncy dance. One pet photo in, a 10-second clip with an original beat out.
- [80s AI Trend: Your 1985 Photo and Home Video](https://www.makeviral.ai/templates/80s-ai-trend): One selfie becomes a faded 1985 snapshot with big hair and a date stamp. Stop there for the photo, or animate it into a 10-second camcorder home video with sound.
- [All AI video templates](https://www.makeviral.ai/templates)

## Related tools and guides

- [GPT Image 2.5 Generator](https://www.makeviral.ai/tools/gpt-image-2-5-generator): OpenAI's image model, in a browser tab. The one to use when the picture has words in it: packaging, app screens, posters and price tags.
- [Z-Image Generator](https://www.makeviral.ai/tools/z-image-generator): The cheap fast one. 1 credit, 1K, 5 to 15 seconds, text prompt only. The right model for backgrounds, b-roll stills and the drafts you throw away.
- [AI Product Photo Generator](https://www.makeviral.ai/tools/ai-product-photo-generator): Describe the product and the scene, or upload the shot you already have. Get back up to four finished images in under a minute, with no studio and no photographer.
- [AI Product Photos for Shopify: Workflow + Costs](https://www.makeviral.ai/blog/ai-product-photos-for-shopify): One clean reference shot, then lifestyle and ad variants from it. Three models compared, 12 copy-pasteable prompts and real credit costs. Try it today.
- [How to Go Viral on TikTok: What Actually Moves Reach](https://www.makeviral.ai/blog/how-to-go-viral-on-tiktok): What TikTok says about recommendations, 15 hooks that hold the first two seconds, a retention checklist and a 14-day plan. Start your first video.
- [Every ai photo generator](https://www.makeviral.ai/ai-photo-generator)

## Put your own photos in the prompt

Text to image or image to image, 1K to 4K, 9:16 first. 2 credits at 1K, about 14 cents on the $49.99 plan for 700 credits.

[Generate with Nano Banana 2](https://www.makeviral.ai/photo?model=nano-banana-2)