# Nano Banana 2: what Gemini 3.1 Flash Image can do

> Nano Banana 2 is Google DeepMind's model for generating and editing images, officially Gemini 3.1 Flash Image. Released on February 26, 2026, it offers near-Pro quality at Flash speed, resolutions from 512 px to 4K and up to 14 references. It's available in the AI2Face generator.

Developer: Google DeepMind · Released: 2026-02-26 · Updated 2026-09-30
URL: https://ai2face.com/models/nano-banana-2

## Key facts

| | |
|---|---|
| Developer | Google DeepMind |
| Official name | Gemini 3.1 Flash Image |
| Release date | February 26, 2026 |
| What it does | Generates and edits images from text and references |
| Resolution | 0.5K, 1K, 2K, 4K (4K square — 4096 × 4096) |
| Aspect ratios | 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9, 1:4, 4:1, 1:8, 8:1 |
| References | Up to 14, including up to 4 photos of people and up to 10 objects |
| Search | Google Web Search and image search |
| Labeling | SynthID watermark on every image |
| In AI2Face | Yes, in the generator |

## What is Nano Banana 2

Nano Banana 2 is a model from the [Nano Banana family](/models/nano-banana) for generating and editing images. Its official name is Gemini 3.1 Flash Image, and its API identifier is `gemini-3.1-flash-image`. Google unveiled it on February 26, 2026: Flash speed plus capabilities that had previously been exclusive to [Nano Banana Pro](/models/nano-banana-pro).

In the Gemini app the model became the default. It replaces Pro in the Fast, Thinking and Pro modes. Google AI Pro and Ultra subscribers can still regenerate an image with Pro via the three-dot menu. In Google Search the model works in AI Mode and Lens. For developers it's open in AI Studio and the Gemini API, but only with a paid key.

Google has no separate “nano banana 2” website. Domains with that name belong to third-party services.

## What Nano Banana 2 can do

- **Resolution from 512 px to 4K.** A 4K square is 4096 × 4096. The 0.5K step exists only in this model of the family.
- **Elongated frames.** Besides the usual 1:1, 16:9 and 9:16, the documentation lists 1:4, 4:1, 1:8 and 8:1 — a banner, a story cover, a wide panorama.
- **Up to 14 references.** Per the API table: up to 4 photos of people to keep their likeness and up to 10 images of objects. One request can combine a person, clothing and an object into a single scene.
- **Search during generation.** The model looks at both the web and Google Images to draw a specific place or thing more accurately. It doesn't insert photos of real people from search — the API limitations say so explicitly.
- **Text and diagrams.** The announcement highlights legible lettering, translating text in images and infographics. Short quoted captions come out more reliably than paragraphs: see the poster and diagram examples in the gallery.
- **Video as input.** Among the family's models, only Gemini 3.1 Flash Image accepts video as input. None of them accept audio.

![Chalkboard menu on a brick wall: on the left the lettering in English, on the right the same lines in Russian](https://ai2face.com/_astro/translate-text.D6M8C0Tp_Z2ts4Lm.jpg "Translating text in Nano Banana 2: an English menu translated into Russian with one prompt. The handwriting, chalk drawings, board and background were preserved.")

## What Nano Banana 2 is good for

Choose Nano Banana 2 when you need a working image quickly, without a separate “studio” pass in Pro.

- **Portraits and series with one face.** Up to four people in one request. For dozens of shots with one character, the AI2Face character builder is more convenient: the face is stored separately and placed into new scenes.
- **Product shots and product cards.** A reference of the packaging or mug plus a text description of the background.
- **Posters and infographics.** Short text in quotes, a clear grid, step captions.
- **Wide and tall formats.** A 16:9 cover, a 9:16 story, an 8:1 banner — no manual cropping of a square.
- **Scenes where “what it really looks like” matters.** Image search helps with architecture and objects. For a legally precise fact, Google recommends Pro: its announcement keeps Pro for tasks where maximum factual accuracy matters most.

## Nano Banana 2 and its alternatives

| Criterion | Nano Banana 2 | [Nano Banana](/models/nano-banana) | [Nano Banana Pro](/models/nano-banana-pro) |
|---|---|---|---|
| Official name | Gemini 3.1 Flash Image | Gemini 2.5 Flash Image | Gemini 3 Pro Image |
| Released | Feb 26, 2026 | Aug 26, 2025 | Nov 20, 2025 |
| Resolution | 0.5K, 1K, 2K, 4K | 1K | 1K, 2K, 4K |
| References | up to 14: 4 people and 10 objects | best 1–3 | up to 14: 5 people, 6 objects, 3 styles |
| Search | web and images | no | web |
| Elongated 1:8 and 8:1 frames | yes | no | no |
| API price, 1K | $0.067 | separate, legacy model | $0.134 for 1K and 2K |
| When to use | most tasks | quick 1K edits | complex layouts and precise text |

Prices are the standard Gemini API tier as of September 30, 2026, not batch. Nano Banana 2 has no free API tier: the pricing page says “Not available” in the Free Tier column. 0.5K costs $0.045, 2K $0.101, 4K $0.151.

## How to write prompts for Nano Banana 2

1. Describe the whole scene: who, where, light, lens. “A man in a gray sweater by a wet window at sunset, 85mm lens” beats a list of tags.
2. Put lettering in quotes and state its role: headline, subtitle, small line. Use the language the lettering should be read in.
3. For a diagram, name the number of steps and the labels. The phrase “all text spelled correctly” reduces typos, but still proofread paragraphs.
4. Give references roles: “the first photo is the face, the second is the clothing, the third is the object.” Without roles the model decides for itself what to transfer.
5. Change one thing at a time. The model is built for conversation: “make it evening”, “move the headline down”.

![Three references — a man, a mustard corduroy jacket and a black mug — and the final shot of him wearing the jacket and drinking from the mug in a kitchen](https://ai2face.com/_astro/references.DRu0FAHP_Z107Efq.jpg "Building a scene from three references in Nano Banana 2: on the left the person, jacket and mug, on the right the result. Each photo has its own role in the prompt: “the first is the person, the second is the jacket, the third is the mug”.")

The full prompts for each shot are in the gallery. The techniques we use every day are collected in the article [Nano Banana prompts](/blog/nano-banana-pro-prompts).

## How to use Nano Banana 2 in AI2Face

Google's official services aren't available in every country: Google publishes a country list. In AI2Face the model is available in the generator, and the cost is shown before you generate.

1. Sign up. New accounts get welcome credits. What exactly is free at Google and where the limits are is covered in the guide [Nano Banana free](/blog/nano-banana-besplatno).
2. Open the generator and choose Nano Banana 2.
3. Write a prompt, upload references if needed, set the aspect ratio and resolution.
4. The cost in credits is shown before you start. We don't quote a number of credits here: it changes in the model catalog.

## Limitations

- Search doesn't insert photos of real people from the internet.
- The API documentation limits face likeness to four characters and objects to ten. The announcement's marketing wording (“up to five characters and up to 14 objects”) doesn't match the API table.
- Long text in an image needs proofreading. The model does better if you write the text separately first and then paste it into the image request.
- Every API image carries a SynthID watermark. On announcement day Google promised to add C2PA verification to the Gemini app, but gave no date.
- Google doesn't publish generation times. We don't carry over figures like “under 10 seconds” or “4K in a minute” from reviews.
- Google considers the original Nano Banana deprecated and advises developers to move to [Nano Banana 2 Lite](/models/nano-banana-2-lite), not back to the first version. Lite is a separate, cheaper 1K-only model.

## Conclusions

Nano Banana 2 is the default model if you don't know where to start. It covers portraits, product shots, posters and unusual frames, looks up visual references and accepts up to 14 references. Switch to Pro when you need a complex layout, more faces in the frame or a separate style reference. The simplest way to compare both versions on the same prompt is AI2Face: the welcome credits are enough.

## Generation examples

- Portrait of a freckled man by a rainy café window at sunset
  Prompt: `Photorealistic close-up portrait of a 32-year-old man with freckles and a short beard, wearing a charcoal knit sweater, standing by a rain-streaked cafe window at golden hour, soft rim light, 85mm lens, natural skin texture, candid half-smile, no text, no logos`
- Minimalist Night Jazz poster with a saxophone and an orange circle
  Prompt: `Minimalist concert poster. Large bold headline text "NIGHT JAZZ", subtitle "Lisbon, 14 June", small line "Doors at 20:00". Deep navy background, one saxophone drawn in thin cream lines, a warm orange circle, clean grid, crisp legible typography, no other text`
- Four-step infographic about how coffee is made
  Prompt: `Clean flat infographic titled "How coffee is made". Exactly four panels in a row, labels only: "1 Harvest", "2 Drying", "3 Roasting", "4 Brewing". Simple illustrations of cherries, drying beds, a roaster and a pour-over. Soft beige background, dark brown text, no other words, all text spelled correctly, no logos`
- Black ceramic mug on a travertine block by a window
  Prompt: `Commercial product photo of a matte black ceramic mug on a travertine plinth, soft morning window light, subtle shadow, dried pampas grass blurred behind, minimal interior, sharp detail, no text, no logos`

## FAQ

### What is Nano Banana 2 and when was it released?

It's Google DeepMind's image generation model. Its official name is Gemini 3.1 Flash Image. Google introduced it on February 26, 2026 as a fast replacement that takes over Pro's capabilities.

### Is Nano Banana 2 the same as Gemini 3.1 Flash Image?

Yes. Nano Banana 2 is the product name, and Gemini 3.1 Flash Image is the model in the API (gemini-3.1-flash-image).

### What is the maximum resolution of Nano Banana 2?

4K, which is 4096 × 4096 pixels for a square. Besides 1K, 2K and 4K, the model can output 512 px (0.5K) — Pro has no such step.

### How many references can I upload to Nano Banana 2?

Up to 14. According to the API documentation, that's up to 4 photos of people to keep their likeness and up to 10 images of objects. Google's announcement mentions five characters — for setting up a generator we go by the API table.

### How is Nano Banana 2 different from Pro?

Nano Banana 2 is faster and cheaper in the API, supports 0.5K and the elongated 1:8 and 8:1 frames, and searches both the web and images. Pro is more precise on complex layouts and accepts up to 5 photos of people plus separate style references.

### Where is the official Nano Banana 2 website?

There's no separate website. The model runs in the Gemini app, Google Search, AI Studio and the Gemini API. These services aren't available in every country. In AI2Face the model is available in the generator, and the cost is shown before you generate.

## Sources

- https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/
- https://ai.google.dev/gemini-api/docs/image-generation
- https://ai.google.dev/gemini-api/docs/pricing
- https://blog.google/innovation-and-ai/technology/developers-tools/build-with-nano-banana-2/