Nano Banana 2 Lite API Guide

Who Should Use the API

This guide is for developers planning an integration that generates or edits images with Nano Banana 2 Lite — Google's gemini-3.1-flash-lite-image model, publicly documented as Gemini 3.1 Flash-Lite Image. It walks through the model's official parameters and the async submit-then-poll pattern this product itself uses via Replicate, so you can plan a comparable integration.

Official Model Code & Capability Summary

The model is officially documented as gemini-3.1-flash-lite-image (Gemini 3.1 Flash-Lite Image) and is also available on Replicate as `google/nano-banana-2-lite`. It accepts a text prompt and, optionally, one or more reference images, and returns a generated or edited image at Native 1K (1024px). nanobanana2lite.me is an independent third-party service and is not affiliated with Google.

Request Parameters

Example request (Replicate, placeholder token)

curl -s -X POST https://api.replicate.com/v1/models/google/nano-banana-2-lite/predictions \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "prompt": "studio product photo of a ceramic mug on a marble counter",
      "aspect_ratio": "1:1",
      "output_format": "jpg"
    }
  }'

Aspect Ratio and 1K Output Notes

Nano Banana 2 Lite generates images at native 1K (1024px) only — there is no higher-resolution output tier. No parameter exists to request a higher-resolution tier from this model — if your product needs output above native 1K, that has to come from a separate upscaling step or a different model tier. See the Generation Workflow docs for the full per-aspect-ratio output reference on this site.

Error Handling & Cost Planning

Plan for asynchronous failures, not just request-level errors: a prediction can accept your input and still resolve to a failed status later, for example if a referenced image URL is invalid. Design your integration so that a failed status triggers a clear user-facing message and, if you charge users per generation, an automatic credit refund — this product refunds credits automatically whenever a generation resolves to failed, rather than only refunding on manual request. For cost planning, this product prices text-to-image and image-to-image separately in its own credit system (2 and 4 credits respectively) — your own per-call cost from the underlying provider is a separate, usage-based figure to confirm directly with your provider account.

Integration Checklist

• Confirm you are calling the correct model identifier and have valid API credentials before going live.
• Implement polling with a reasonable timeout and backoff, since this is an asynchronous API.
• Handle the failed status explicitly, including any user-facing refund or retry messaging.
• Validate reference image URLs are publicly reachable before submitting an image-to-image request.
• Cap the number of reference images and edit rounds per request to match official guidance.
• Log the returned task/prediction id so failed requests can be traced back to a specific submission.