Nano Banana 2.1 is Google’s image generation and editing model, model ID gemini-nano-banana-2.1. Google Cloud says it is optimized for multimodal image generation and editing with good cost performance. OpenRouter places it on the Flash tier, calling it the successor to Nano Banana 2 and Nano Banana Pro, with a release date of October 6, 2026.
If you use Nano Banana 2, 2.1 is a refinement of the same tier: steadier long batches, closer reference matches, more precise chat edits. It stays on the Flash path, speed first and predictable cost, for daily image batches.
What Nano Banana 2.1 is
Nano Banana 2.1 is the newest image model in Google’s Nano Banana line, written in Google Cloud docs as Gemini Nano Banana 2.1 and optimized for multimodal image generation and editing. The model ID is gemini-nano-banana-2.1, which you call when integrating.
It is a Flash-tier model, not Pro. OpenRouter states it is Google’s Flash-tier image model that succeeds Nano Banana 2 and Nano Banana Pro. Google gives no specific price, only good cost performance; OpenRouter lists input 1.50 dollars and output 30 dollars per million tokens, but that is not the official pricing page.
What changed from Nano Banana 2
Nano Banana 2.1 follows Nano Banana 2 (Gemini 3.1 Flash Image, February 26, 2026). The timeline runs: original Nano Banana (August 26, 2025), Pro (November 20, 2025), Nano Banana 2 (February 2026), then 2.1 on October 6, 2026.
OpenArt calls 2.1 the line’s latest fast image model and lists three gains: steadier long batches, closer matches to your references, precise chat edits. The daily win is batch stability: from one reference you can get up to 40 style-consistent images, and 2.1 holds that cross-batch consistency more reliably.
Where you can use Nano Banana 2.1
Today there are three kinds of places to use it.
- Official route: Google Cloud docs are the authoritative source for identity and capabilities, and the model can be tried in Agent Studio and viewed in Model Garden.
- Aggregated API: OpenRouter lists it as a callable model, marked Flash-tier with a release date of October 6, 2026, served by Google AI Studio.
- Vertical platforms: fal.ai shipped the endpoint
google/nano-banana-2.1, text-to-image, up to 4K, adjustable aspect ratio, optional web search grounding, with a commercial use note. OpenArt offers a 2.1 page with up to 40 consistent images from one reference and chat editing.
The difference is billing and surrounding features; the model itself is the same gemini-nano-banana-2.1.
Resolution, references and limits
Output is 1K, 2K or 4K with fifteen aspect ratios: 1:1, 3:2, 2:3, 3:4, 1:4, 4:1, 4:3, 4:5, 5:4, 1:8, 8:1, 9:16, 16:9, 21:9, 9:21. Each prompt takes up to 14 image references.
The officially documented features that matter for output work:
- Multi-turn edits: keep working on the same image across rounds.
- Video input: generate an image from video.
- Interleaved images and text: the answer carries both.
- Virtual try-on: swap how a garment sits on a person.
- C2PA Content Credentials: finished images carry provenance metadata.
- Google and image search grounding: call it when real-world facts matter.
Two limits are written just as plainly: portrait generation is not supported, so do not generate faces, and structured output is not supported, so do not expect JSON.
Cases: two prompts you can copy
Both cases keep a matte ceramic water bottle as the subject across scenes. See the image, then copy the prompts.
Case one: from one product reference, generate a set of scene images. The fence holds the prompt to copy.
Use the attached photo as the only product reference. Keep the bottle exactly as it is: shape, matte glaze and proportions. Generate it standing on a round studio pedestal under soft diffused light.Expected: the model keeps the bottle’s shape, matte glaze and proportions on a studio pedestal, only the scene and light change. Swap the last sentence to move it to a kitchen, picnic or restaurant; OpenArt can pull up to 40 consistent results from one reference.
Case two: a chat edit that changes only light and background. Send one more sentence naming only the change.
Change only the lighting to a warm evening mood and the background to a dark wooden shelf. Keep the bottle exactly the same: shape, color, surface and proportions. Do not regenerate the rest.Expected: the bottle does not move; only the light turns warm evening and the background becomes a dark wooden shelf. That is the official precise chat edit: name only the one change and lock the rest.
How to write prompts that hold up
Separate what changes from what stays locked. OpenArt’s advice: write subject, scene and light plainly, name details that must stay stable across batches, and when editing say only the one change. The fences show this: case one fixes the reference and subject, case two pins the rest with “only” and “Do not regenerate the rest.” Repeat “keep the bottle unchanged” each round to curb drift.
Resolution and ratio are call parameters, not written into the prompt. Turn on search grounding for real-world facts, but do not expect structured output.
For more, the Nano Banana Pro prompt guide goes deeper, how to use Nano Banana Pro gives entry points, the Nano Banana Pro photo editor covers editing, and Nano Banana Pro pricing tells listed price from official pricing.
FAQ
Is Nano Banana 2.1 free, and how is it priced? Google Cloud gives no specific price, only good cost performance. OpenRouter lists input 1.50 dollars and output 30 dollars per million tokens, but that is not the official pricing page.
How is it related to Nano Banana 2 and Pro? 2.1 follows Nano Banana 2 (Gemini 3.1 Flash Image, February 26, 2026); OpenRouter says on the Flash tier it succeeds both. They are different generations of one line, not the same model.
Which resolutions and references does it support? Output is 1K, 2K or 4K with fifteen aspect ratios; each prompt takes up to 14 image inputs and supports multi-turn edits, video input, interleaved images and text, and virtual try-on.
Can it be called through an API? Yes. OpenRouter and fal.ai both provide callable endpoints (fal.ai is google/nano-banana-2.1), with text-to-image, up to 4K, adjustable aspect ratio and optional search grounding.