# Nano Banana (Gemini 2.5 Flash Image)

Nano Banana (Gemini 2.5 Flash Image) is a native image generation and editing model that brings Gemini's world knowledge directly into the image output pipeline, enabling character-consistent storytelling, precise natural-language edits, and multi-image fusion without separate image-processing services.

- **Model ID:** `google/gemini-2.5-flash-image`
- **Type:** chat
- **Providers:** google, vertex
- **Context window:** 32,768
- **Maximum output tokens:** 65,535
- **Pricing:** $0.3/1M input tokens, $2.5/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/gemini-2.5-flash-image

## Supported parameters

- **Parameters:** `temperature`
- **Operations:** Not reported
- **Input modalities:** `text`, `image`
- **Output modalities:** `text`, `image`
- **Aspect ratios:** Not reported
- **Resolutions:** Not reported
- **Durations:** Not reported
- **Frame rates:** Not reported
- **Maximum outputs per request:** Not reported
- **Maximum input images:** Not reported
- **Input image formats:** Not reported
- **Maximum input image size:** 7 MB

## Example

```ts
import { generateText } from 'ai'

const result = await generateText({
  model: 'google/gemini-2.5-flash-image',
  prompt: 'Render a picture of a red balloon.',
});
```

## About

Google introduced Nano Banana (Gemini 2.5 Flash Image) (internally codenamed nano-banana) on August 26, 2025 as a direct response to developer feedback on the earlier native image generation in Gemini 2.0 Flash. Users valued 2.0 Flash's low latency and ease of use but requested higher image quality and more capable creative control.

Four capabilities define the model. First, character consistency: it can place the same character or object into different environments, generate a product from multiple angles in new settings, or produce consistent brand assets across a series of prompts while preserving subject appearance. Second, prompt-based image editing: you use natural language to perform targeted local edits (blurring a background, removing a stain, altering a pose, colorizing a black-and-white photo) in a single call. Third, native world knowledge: unlike prior image generation models that excelled at aesthetics but lacked semantic grounding, Nano Banana (Gemini 2.5 Flash Image) draws on Gemini's world knowledge to interpret hand-drawn diagrams, answer questions grounded in real-world understanding, and follow complex editing instructions in one step. Fourth, multi-image fusion: the model accepts multiple input images and merges them, enabling product placement into new scenes, room restyling from a reference texture, and image-to-image blending.

All images created or edited with Nano Banana (Gemini 2.5 Flash Image) include an invisible SynthID digital watermark for downstream identification of AI-generated or AI-edited content.

## What to consider

Image generation from Nano Banana (Gemini 2.5 Flash Image) is priced per image at N/A. Other input and output modalities follow Gemini 2.5 Flash rates. See the pricing panel on this page for current rates.

## When to use

### Best For

- **Visual storytelling and character consistency**, campaigns, comics, or narrative applications that require the same character to appear coherently across multiple distinct images
- **Automated product photography**, generating a product catalog from a single reference image, showing items in multiple settings and angles at scale
- **Prompt-driven photo editing**, building user-facing editing tools that accept natural language instructions to perform precise, targeted modifications to uploaded images
- **Multi-image composition workflows**, fusing product images into lifestyle scenes, restyling interiors with reference textures, or merging source materials into a single photorealistic output
- **Education and knowledge-grounded visuals**, generating diagrams, illustrations, or annotated visuals that require semantic understanding of real-world concepts rather than purely aesthetic generation

### Consider Alternatives When

- Your application needs only text output and image generation would add unnecessary cost and complexity
- You require video generation rather than still image output
- The application is latency-sensitive and purely text-based, where a standard Gemini 2.5 Flash model is more appropriate
- You need embedding or retrieval capabilities, which require a dedicated embedding model architecture

## Best for

- **Visual storytelling and character consistency:** campaigns, comics, or narrative applications that require the same character to appear coherently across multiple distinct images
- **Automated product photography:** generating a product catalog from a single reference image, showing items in multiple settings and angles at scale
- **Prompt-driven photo editing:** building user-facing editing tools that accept natural language instructions to perform precise, targeted modifications to uploaded images
- **Multi-image composition workflows:** fusing product images into lifestyle scenes, restyling interiors with reference textures, or merging source materials into a single photorealistic output
- **Education and knowledge-grounded visuals:** generating diagrams, illustrations, or annotated visuals that require semantic understanding of real-world concepts rather than purely aesthetic generation

## Consider alternatives

- **Text-only output needed:** Image generation would add unnecessary cost and complexity
- **Video generation required:** Still image output is not sufficient and a video model is the right fit
- **Latency-sensitive text pipelines:** A standard Gemini 2.5 Flash model is more appropriate for purely text workflows
- **Embedding or retrieval workloads:** A dedicated embedding model architecture is required

## Frequently asked questions

### What is the per-image cost for Nano Banana (Gemini 2.5 Flash Image)?

Current pricing is shown on this page. AI Gateway routes across providers, and rates may vary by provider.

### Can the model maintain a character's appearance across multiple generated images?

Yes. Character consistency is one of the four headline capabilities described. The model can reproduce the same character or object in different environments, angles, and settings while preserving visual identity.

### What types of prompt-based edits can the model perform?

You can perform edits like blurring backgrounds, removing subjects from scenes, adding or replacing elements, changing colors, and applying style transfers. The model handles these through natural language prompts combined with input images.

### How does multi-image fusion work?

The model accepts multiple images as input and can merge them in a single prompt, for example, placing a product into a new scene, restyling a room using a reference texture or color scheme, or blending two source images together.

### What is SynthID and are outputs watermarked?

SynthID is Google's invisible digital watermark technology. All images created or edited with Nano Banana (Gemini 2.5 Flash Image) include a SynthID watermark that allows them to be identified as AI-generated or AI-edited.

### What makes this model's world knowledge capability distinct from prior image generation models?

Previous image generation models excelled at aesthetics but lacked deep semantic understanding. Nano Banana (Gemini 2.5 Flash Image) draws on Gemini's world knowledge to interpret hand-drawn diagrams, reason about real-world questions, and follow complex multi-step editing instructions in a single generation step.

### What are the known limitations?

Known limitations at preview launch include long-form text rendering within images, character consistency reliability, and factual accuracy of fine image details. Google is actively improving these areas.

## Links

- [Model page](https://vercel.com/ai-gateway/models/gemini-2.5-flash-image)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-image)
- [Provider pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing)