Nano Banana 2 API: Pricing, Model Name and Cost per Image

If you are pricing Nano Banana 2 for an app, a store catalog or a content pipeline, the answer fits in one line: the API model name is gemini-3.1-flash-image, and Google charges $0.045 to $0.151 per generated image, depending on resolution. The rest of this article is the detail that prevents a surprise invoice: what each resolution costs, how the half-price batch rate works, which extra fees can appear, and how the model stacks up against the rest of the Nano Banana family. Every price comes from Google's own pricing and image generation pages, checked in October 2026, and anything that comes from somewhere else is labeled.

💡 Rule of thumb: budget about $0.07 per image at 1K, $0.10 at 2K and $0.15 at 4K. Halve each figure for batch jobs.

What the Model Is Called

Nano Banana 2 is the nickname. In code you call gemini-3.1-flash-image, and Google's pricing page lists it as generally available. The official docs show the ID with no preview suffix, which matters because older tutorials and some gateways still use a -preview name. If a snippet you found online fails, check the model string first.

Model IDs across the family

Google ships four related image models, and the names are easy to mix up. This table maps each nickname to its API ID, with a link to the matching page on PicassoIA.

The shutdown date matters if you maintain older code. Google's pricing page marks gemini-2.5-flash-image as deprecated with a shutdown date of October 2, 2026, so any project still pointing at it needs a new model string.

Names you will see on resellers

Third-party gateways keep their own labels. OpenRouter, for example, lists the model as google/gemini-3.1-flash-image-preview, a leftover from the preview period. A reseller's model string is not Google's string, so copy the exact ID from the platform you are paying. If you call Google directly, use gemini-3.1-flash-image.

Cost per Image by Resolution

Google bills image output by the token, then publishes a flat price per image at each size. In practice you can treat it as four price points instead of a running meter.

Standard rates

Each step up costs more than it looks. Going from 1K to 2K adds about 51%, and going from 1K to 4K multiplies the bill by roughly 2.25. One 4K render costs the same as two and a quarter renders at 1K.

Batch rates

Google's Batch API takes 50% off both input and output tokens, which lands on these prices per image:

The trade is speed. Batch work is asynchronous, so it suits catalogs, backfills and overnight jobs rather than a button a customer is waiting on.

How token billing works

Behind the flat prices sits token billing: $0.50 per million input tokens (text and images) and $60 per million output tokens (the generated image). Batch halves both, to $0.25 and $30. Your prompt barely registers, since a 200-token prompt costs about $0.0001. The image output is what you pay for, so the resolution you request decides almost the whole bill.

Charges beyond the image

• No free API tier. Google lists no free tier for this model, so testing costs money from the first call.
• Search grounding. If you let the model ground an image in Google Search, you get 5,000 free search requests per month, shared across all Gemini 3.x models, then pay $14 per 1,000 requests. If every call used one search request, that would add $0.014, about 21% on top of a 1K image.
• Reference images. They are billed as input tokens at the $0.50 per million rate.
• Reseller rates. Resale platforms advertise their own flat per-image prices. In the listings checked for this article they ranged from about $0.03 to $0.065, but those are advertised figures, not verified ones, so read the terms before you build a budget on them.

How It Compares With Its Siblings

Here is the whole family side by side, using Google's published per-image prices.

When Lite is enough

Nano Banana 2 Lite costs $0.0336 at 1K, about half of the standard model, but it stops at 1K and does not support grounding. It fits thumbnails, drafts, mood boards and bulk variations where you will never need 2K, 4K or live search data.

When Pro still earns its price

Nano Banana Pro costs $0.134 at 1K and 2K and $0.24 at 4K. At 1K it is exactly double Nano Banana 2. At 2K the cheaper model saves $0.033 per image (25%), and at 4K it saves $0.089 per image (37%). Pay the Pro premium only when a side-by-side test on your own prompts shows a difference you can actually see.

Price is not the only filter. If your prompts do not need Google's grounding or reference-image fusion, run the same brief through GPT Image 2, Seedream 5 Pro and Flux 2 Pro. Each has its own page on PicassoIA, and ten minutes of comparison beats guessing.

Calling It From Code

Google's current docs show the model through the Interactions API in Python. Install google-genai, put your API key in the environment, then run:

from google import genai
import base64

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-3.1-flash-image",
    input="Create a picture of a nano banana dish in a fancy restaurant with a Gemini theme",
)

with open("generated_image.png", "wb") as f:
    f.write(base64.b64decode(interaction.output_image.data))

Copy Prompt

The image comes back base64 encoded, so decode it before writing to disk, as the last two lines do. Google's docs also describe a thinking_level setting inside generation_config with two levels, minimal and high. Check how it affects latency and cost before you switch it on at scale.

Resolutions and aspect ratios

The model accepts four sizes: 0.5K (512px), 1K, 2K and 4K. Supported aspect ratios are 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9 and 21:9. Lite is limited to 1K.

Reference images and watermarks

You can pass up to 10 object images for high-fidelity inclusion and up to 4 character images for consistency across scenes. Every generated image carries a SynthID watermark, Google's provenance marker, so plan for that in any print or stock workflow.

Monthly Cost Examples

Three workloads, priced at list rates. All figures ignore tax and input tokens, which add roughly a cent per hundred images with short prompts.

A small shop with 300 product shots

A ceramics seller refreshing a 300-item catalog at 1K pays 300 × $0.067 = $20.10 on the standard endpoint, or $10.20 through batch. Rerunning the entire catalog every month still stays near twenty dollars.

A content team at 2K

A team producing 2,000 blog and social images a month at 2K pays 2,000 × $0.101 = $202 standard, or $100 through batch. If a quarter of the images need one retry, add 25% and the standard bill becomes $252.50.

A 4K print run

For 500 large prints at 4K, the math is 500 × $0.151 = $75.50 standard, or $38 through batch. The same 500 images at 1K would cost $33.50, so the jump to 4K adds $42. That upgrade makes sense for print and large displays, and rarely for anything viewed on a phone.

Four Mistakes That Inflate the Bill

1. Defaulting to 4K. It costs about 2.25 times a 1K image. Draft at 1K or 2K and render only the final picks at 4K.
2. Leaving search grounding on everywhere. After 5,000 free requests a month, each extra 1,000 costs $14. Switch it on only for prompts that need current facts.
3. Running bulk jobs on the standard endpoint. If nobody is waiting on the result, batch cuts the price in half.
4. Pointing code at a retired ID. gemini-2.5-flash-image has a listed shutdown date of October 2, 2026. Use gemini-3.1-flash-image for new work.

💡 Tip: log the resolution and the model ID next to every generated image. When the bill arrives, you can trace any spike to a specific job in seconds.

Run Nano Banana 2 on PicassoIA

Code is not the only way in. Nano Banana 2 on PicassoIA runs in the browser, with no API key and no billing setup. The model page states there are no per-generation credits or usage quotas, which makes it a cheap place to settle a prompt before you pay for API calls. It is a browser tool, so for programmatic access you still call Google's API as shown above.

1. Open the model page. Start from the Nano Banana 2 page linked above.
2. Write the prompt. Name the subject, the setting and the light. For longer briefs, draft the wording with Gemini 3.5 Flash for speed, or Gemini 3.1 Pro when the brief is long and detailed.
3. Pick an aspect ratio. There are 15 presets, including 16:9, 9:16, 4:5 and 1:1. The default, match_input_image, follows your reference photo when you upload one.
4. Choose the resolution. Options are 1K (the default), 2K and 4K. Test at 1K first, since higher resolutions take longer to generate.
5. Add reference images. Upload up to 14 to blend looks or keep a character consistent.
6. Decide on grounding. The Google Search toggle ties the image to current information, and the Image Search toggle pulls web images in as visual context.
7. Set the output format. JPG is the default, PNG is available.
8. Generate, then refine. Type a follow-up instruction such as "make the background darker" instead of rewriting the whole prompt.

💡 Budget trick: settle composition, wording and aspect ratio on PicassoIA, then send only the finished prompt to the paid API. Nano Banana 2 Lite is also there for quick drafts. It has no resolution setting, which keeps rough passes simple, and you can browse super resolution and other tools on the all models page.

Try It Yourself on Picasso IA

The numbers are easy to read and harder to feel, so run your own test. Take one prompt from your real project, open Nano Banana 2 on Picasso IA, and generate it at 1K, 2K and 4K. Compare the three side by side, then multiply the gap by the number of images you plan to make each month. That single experiment tells you whether 1K is plenty, whether Lite fits your volume, and whether Pro is worth the extra spend. Experiment freely, refine the wording with Gemini, and save the paid API calls for the version you have already approved.