Google's Gemini Nano Banana 2.1 supports text-to-image generation and reference-image editing. APIMart provides official and Ext routes with the same output resolutions, but different billing and limits.

Generated image links expire 72 hours after completion. Download your images promptly.
Transparent pricing with no hidden fees. Pay only for what you use.
* Actual costs are subject to final output.
* gemini-nano-banana-2.1: Image-output estimate only. Input and reasoning are billed separately; the final charge is based on actual usage.
* If Ext is unavailable, allow fallback to the official model. A fallback task uses official token-based billing, so the final charge may differ from the Ext estimate.
Turn prompts and reference images into visual drafts. Choose 1K, 2K or 4K, set an aspect ratio, and use the official or Ext route for your workflow.
Supported image inputs, output settings and task workflow
Practical ideas for creative, commerce and content teams
Compare every Nano Banana model on APIMart. One API key calls all of them, billed per image.
| Model | Base model | Best for | Resolution |
|---|---|---|---|
| Nano Banana API | Gemini 2.5 Flash Image | Character consistency, multi-image blending and conversational editing | 1K |
| Nano Banana Pro API | Gemini 3 Pro Image | Professional 4K images, complex layouts and multi-step planning | 1K / 2K / 4K |
| Nano Banana 2 API | Gemini 3.1 Flash Image | Fast, cost-efficient generation from 0.5K to 4K | 0.5K – 4K |
| Nano Banana 2 Lite API | Gemini 3.1 Flash Image | Fast 1K images and editing, one image per request with up to 14 reference images | 1K |
| Gemini Nano Banana 2.1 APIThis page | gemini-nano-banana-2.1 | Latest version: text-to-image and reference-image editing in 14 aspect ratios | 1K / 2K / 4K |
Routes, parameters, billing and result handling
gemini-nano-banana-2.1 uses the Google official route, while gemini-nano-banana-2.1-ext uses a third-party route. Both use the same upstream model, but supported aspect ratios depend on the route and resolution; billing, image count and reference-size limits also differ.
The official route supports n from 1 to 4. Ext currently supports only n=1. Use separate Ext tasks or the official route when you need several images.
image_urls accepts publicly accessible HTTP(S) URLs or base64 data images. PNG, JPEG and WEBP are recommended. Official limits each reference to 20 MB; Ext allows 30 MB each and 50 MB combined. Upstream limits still apply even though the platform sets no reference-count cap.
Choose 1K, 2K or 4K and auto, 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 or 8:1. Both the official route and Ext support all these ratios at every resolution, including 1:4, 4:1, 1:8 and 8:1 at 2K and 4K. Defaults are 1K and auto. This model does not support 0.5K.
Official billing includes actual input tokens and output tokens, including image output and thinking. An image-output estimate is not the full task price. Ext is priced per image by resolution. Check the platform price list; the completed task's cost is the final charge.
This optional Ext setting is off by default. If the third-party route fails and the task completes through the official route, official token-based billing applies instead of Ext per-image pricing.
POST /v1/images/generations returns a task_id. Query GET /v1/tasks/{task_id} until the task completes or fails. Completed image links are returned in result.images[].url; include every URL when requesting multiple images.
Image links expire 72 hours after completion; use expires_at and save the files promptly. A task marked failed is fully refunded and reports cost=0. A network error while checking status does not by itself mean the task failed.
Explore more models in the same category.

Seedream 5.0 Flash
Seedream 5.0 Flash is ByteDance's fast, lower-cost text-to-image model. It supports single-image and multi-reference editing with up to 10 references and generates 1K, 1.5K or 2K images, or exact pixel sizes, with strong text rendering.

GPT Image 2.5 Flare
GPT Image 2.5 Flare is OpenAI's speed-focused GPT Image 2.5 model for image generation and editing. It supports up to 16 reference images, inpainting with transparent PNG masks and 15 aspect ratios.

GPT Image 2.5 Sunburst
GPT Image 2.5 Sunburst is an OpenAI image model for generation and editing, focused on precise edits for detail-sensitive creative work. Supports up to 16 reference images and PNG, JPEG or WebP output.

Grok Imagine Image Quality
Grok Imagine Image Quality is the quality-focused version of xAI's Grok image model for photorealistic text-to-image and image editing, with up to three reference images.