Answer first: Nano Banana 2 is Google’s Gemini 3.1 Flash Image family, not Gemini 2.5 Flash Image. For the Gemini API, use the current preview model ID shown in Google’s documentation and calculate cost from the live pricing row for the selected resolution.
Product Name and API Model
| Name | Current meaning |
|---|---|
| Nano Banana | Earlier Gemini 2.5 Flash Image generation model |
| Nano Banana Pro | Gemini 3 Pro Image family |
| Nano Banana 2 | Gemini 3.1 Flash Image family |
| Current API model | gemini-3.1-flash-image |
Do not copy a 2.5 model ID, token rate, or resolution assumption into a Nano Banana 2 integration.
Pricing Boundary
Google’s current pricing page charges the API model for input tokens and image-output tokens. Output cost depends on resolution, so a single “price per image” without a resolution and date is incomplete.
Build a cost record with:
- model ID;
- output resolution;
- text and image inputs;
- successful and failed calls;
- output images returned;
- retries;
- the price page checked at billing time.
Google can change models, limits, and pricing. The billing console and current documentation are the operational source of truth.
A Safe API Evaluation
Run a small test before production:
- Pin the exact model ID.
- Use licensed or synthetic reference images.
- Test text rendering, edits, subject consistency, and composition separately.
- Record token usage and resolution from responses.
- Review safety and provenance behavior.
- Set a project budget and rate limit.
- Monitor the deprecations and model lifecycle pages.
Do not treat a consumer app’s free access or subscription as evidence of free API output.
Measure Cost Per Accepted Image
Use 20 representative briefs. For each one, count:
- generated images;
- accepted images;
- correction turns;
- human editing minutes;
- API input and output charges.
Cost per accepted image equals total API and editing cost divided by accepted images. This is more useful than the list price of one successful output.
Access Surfaces Are Separate
Google AI Studio, the Gemini Developer API, Vertex AI, consumer Gemini experiences, and third-party products can expose different models, limits, billing, and data terms. Verify the exact surface instead of assuming the nickname implies identical access.
Provider availability also does not confirm Flowith availability. Check the live Flowith model picker or an official Flowith announcement before describing the model as supported here.
Related Flowith Guides
- Use Nano Banana 2 alternatives when another image workflow may fit.
- Use Nano Banana 2 vs. GPT Image for a same-brief text-rendering test.
- Use Nano Banana SynthID and export FAQ for provenance and output checks.
FAQ
What is Nano Banana 2’s official Gemini API model?
Google’s current image-generation documentation identifies Nano Banana 2 as Gemini 3.1 Flash Image and uses the API model ID gemini-3.1-flash-image.
Is Nano Banana 2 the same as Gemini 2.5 Flash Image?
No. Gemini 2.5 Flash Image is the earlier Nano Banana model. Do not use its model ID or price row as Nano Banana 2 authority.
How is Nano Banana 2 API use billed?
Use the current Gemini API pricing row for the exact model. Input and output are token-billed, and image output cost varies with resolution.
Is Nano Banana 2 automatically available in Flowith?
Provider API availability does not prove availability in Flowith. Check Flowith’s live model picker or official announcement separately.
Official Sources
Sources last checked July 28, 2026: