Creating a Nano Banana API token costs nothing, yet Google lists no free tier for any of its image models. This article lays out the real price per image, what happens when billing is off, and the free ways to try Nano Banana without paying for API calls.
You can create a Nano Banana API token for free in about a minute. You cannot use it to generate images for free. As of October 2026, Google's pricing page lists no free tier for any Nano Banana model, so every image requested through the Gemini API is billed from the first call. That single fact explains why so many developers see a 429 error on a brand new token that looks perfectly valid.
This article separates what is free from what is not. You will find the exact steps to create the token, the model ID and price of every Nano Banana version, three budget examples, the quota rules behind the errors, and the legitimate ways to try the model without paying for API calls. Every figure was checked against Google's documentation in October 2026. Prices change, so confirm them on the official pricing page before you set a budget.
💡 Quick answer: Creating a token is free. Calling Nano Banana through the API needs a billing account. Trying Nano Banana without billing means using an app or a browser tool, not the API.
Nothing. Google's documentation says new users receive a default Google Cloud project and a token automatically after they accept the Terms of Service in Google AI Studio. No payment method is needed to get the token itself. The token only identifies your project, and it carries no image quota of its own, which is the part most people miss.
Think of it as a library card that comes with no books. The card is free, and the shelf behind it, the image models, is paid.
Every Nano Banana model is paid. The price runs from $0.0336 for a 1K image on the lightest model up to $0.24 for a 4K image on the Pro model. Google's text models do come with free tier quotas and daily request caps, and that reputation is why so many tutorials promise a free image API. The promise does not hold for the image models.
The practical result is simple:
The sections below show each path in order.
Google can change this at any time, and free tiers for other model families have moved before. Do not rely on the date of a tutorial. Open the pricing page, find the row for the model ID you plan to call, and read its free tier cell yourself. If that cell ever lists a quota instead of "Not available," the advice in this article changes with it.
Nano Banana is the nickname for Google's Gemini image models. The nickname now spans several generations, and each one has its own model ID, resolution ceiling, and price. Calling the wrong ID is the fastest way to overspend, so the table lists them side by side.
| Nickname | Model ID | Free tier | Price per image |
|---|---|---|---|
| Nano Banana | gemini-2.5-flash-image | None | $0.039 |
| Nano Banana 2 Lite | gemini-3.1-flash-lite-image | None | $0.0336 at 1K |
| Nano Banana 2 | gemini-3.1-flash-image | None | $0.045 at 0.5K, $0.067 at 1K |
| Nano Banana Pro | gemini-3-pro-image | None | $0.134 at 1K or 2K, $0.24 at 4K |
Resolution limits differ by model. Nano Banana 2 Lite renders 1K only. Nano Banana 2 offers 512px (0.5K), 1K, 2K, and 4K. Nano Banana Pro offers 1K, 2K, and 4K. The three newer models accept up to 14 reference images, and every generated image carries a SynthID watermark.
A workable pattern uses two stages. Draft the composition on Nano Banana 2 Lite or at 0.5K until the framing, the pose, and the lighting look right. Then send the one winning prompt, with its reference images, to Nano Banana Pro for the final render. You pay the premium price once per deliverable instead of once per experiment.
GEMINI_API_KEY or GOOGLE_API_KEY, and if both exist, GOOGLE_API_KEY wins.export GEMINI_API_KEY="paste-your-token-here"
Most people can stay in the default project. A second project is only useful when you want separate billing and separate limits, for example one for a client site and one for personal experiments. Remember that quotas belong to the project, so splitting work across projects also splits the bill.
Google's documentation shows this minimal Python call. The client reads the environment variable on its own, so no string appears in the code.
from google import genai
import base64
client = genai.Client()
interaction = client.interactions.create(
model="gemini-3.1-flash-image",
input="A ripe banana on a wooden desk, morning window light",
)
with open("banana.png", "wb") as f:
f.write(base64.b64decode(interaction.output_image.data))
On a project without billing, this is the exact call that fails. On a billed project, it writes banana.png and charges roughly 6.7 cents.
Google tells developers to treat the token like a password. If it leaks, other people can burn your quota and run up charges on your project.
A token without billing behind it can authenticate, but it cannot pay for an image. Developers on Google's own forums report 429 responses when they call a Nano Banana model from a free project, and the message reads like a rate problem even though the real cause is missing billing.
Check these in order:
Three habits waste a day of debugging:
Fix billing first. Then, if the 429 responses continue, treat them as a rate issue and slow the request loop.
Google measures usage in requests per minute, tokens per minute, and requests per day. Models that generate pictures also count images per minute (IPM), a unit that works like tokens per minute. Limits apply to the project, not to the token, so a second token inside the same project adds no capacity.
Google's rate limit page does not print fixed numbers for image models. It points you to AI Studio, where your active limits are listed for each project. Read them there instead of trusting a blog table, including this one.
💡 Tip: Retries are billed like any other call. Add exponential backoff to your code so a burst of 429 responses does not turn into a burst of charges.
| Model and size | Per image | 100 images | 1,000 images |
|---|---|---|---|
| Nano Banana 2 Lite, 1K | $0.0336 | $3.36 | $33.60 |
| Nano Banana | $0.039 | $3.90 | $39.00 |
| Nano Banana 2, 0.5K | $0.045 | $4.50 | $45.00 |
| Nano Banana 2, 1K | $0.067 | $6.70 | $67.00 |
| Nano Banana Pro, 1K or 2K | $0.134 | $13.40 | $134.00 |
| Nano Banana Pro, 4K | $0.24 | $24.00 | $240.00 |
Prices come from Google's pricing page in October 2026.
The third case is where bills surprise people. Every regeneration is a billed image, so a loose prompt costs more than a careful one. Draft at 0.5K or on the Lite model, then render the winner once on Pro.
💡 Tip: Create a budget alert in Google Cloud Billing before your first paid call. It takes two minutes and warns you before a runaway loop does damage.
Google's Gemini app lists Nano Banana image generation among its features across subscription plans. Google's support page does not publish a fixed daily number for free accounts, so the cap you see inside the app is the one that counts, and it can change without notice. The app is manual only. It has no API, so it cannot feed a script or a content pipeline.
Picasso IA runs several Google image models in the browser with no token and no Cloud account. The Nano Banana 2 page says generations carry no credit caps or usage quotas. The same family includes Nano Banana 2 Lite, Nano Banana Pro, and Gemini 2.5 Flash Image.
If you want help writing prompts first, chat models such as Gemini 3.5 Flash draft and refine them in the same place.
A fair test of any route uses three prompts. First, ask for a scene with one short word of text, to see how the model handles lettering. Second, upload a single reference photo and ask for the same person in a new location, to check consistency. Third, edit your own result with a plain sentence such as "move the lamp to the left shelf." A model that passes all three is ready for real projects.
| Route | Needs a token | Needs Google billing | Works in scripts |
|---|---|---|---|
| Gemini API | Yes | Yes | Yes |
| Gemini app | No | No | No |
| Picasso IA in the browser | No | No | No |
The browser route needs a few clicks and zero setup. These steps use the schema published on the model page.
match_input_image, and presets run from 1:1 and 16:9 to 21:9, 9:16, and tall formats such as 1:4.| Setting | Options | Tip |
|---|---|---|
image_input | Up to 14 images | Reuse one reference to keep a character consistent across scenes |
resolution | 1K, 2K, 4K | Draft at 1K, finish at 4K |
output_format | jpg, png | Choose png for graphics with hard edges |
google_search | On or off | Turn it on when the image depends on current events |
image_search | On or off | Adds web images as visual context |
Nano Banana Pro uses a similar form, with a 2K default resolution and a safety filter setting that ranges from strict to permissive. Pick it when the job is a final deliverable.
💡 Tip: Generate three variations at 1K before spending time on a 4K render. Most bad prompts show up in the first thumbnail.
You now know the honest picture: the token is free, the API calls are not, and the browser route skips both problems. If your goal is to see what Nano Banana can do, you do not need a billing account to start. Open Picasso IA, pick Nano Banana 2 or Nano Banana Pro, and write the first prompt that comes to mind.
Try a product shot, a portrait with reference photos, or a scene grounded in today's news. Then compare the results across models on the full model list. Experiment with the settings in the table above, keep the prompts you like, and build a small library of what works. When you are ready to automate, you will already know which model and resolution deserve your API budget. Before you do, return to the price table in this article and run your own numbers with the volume you expect. A rough monthly estimate takes one minute and prevents an unpleasant surprise on the first invoice.
