Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Nano Banana 2.1 became GA October 6, 2026, with ID gemini-nano-banana-2.1.
- It improves text, multiround character consistency and instructions, supporting 1K/2K/4K and wider aspect ratios.
- Standard image output is $0.0336 per 1K image, $0.0504 at 2K and about $0.113 at 4K. Input/text output are separate and no API free tier exists.
- Old gemini-3.1-flash-image is deprecated, but October 7 English docs give no shutdown date. Prepare migration without an immediate rumor-driven switch.
Poster text and iterative editing are the practical focus
The October 6 Gemini API update announces Nano Banana 2.1 GA, succeeding Nano Banana 2/gemini-3.1-flash-image. API GA does not establish simultaneous features in every Gemini app, plan or region.
Google lists visual quality, instruction following, text and multiround character consistency improvements. Evaluate whether people/products remain recognizable through edits and titles appear correctly. These are goals, not model tests performed here.
Hypothetical example: generate a product poster, then widen it, change background and preserve person/product. Continuous editing fits that task, but brand names, prices and promotion conditions still need human checks; generated text is not a product fact.
Lower output rates do not halve total cost
As of October 7, Standard paid image output is $30 per million tokens, equivalent to $0.0336 at 1K, $0.0504 at 2K and about $0.113 at 4K. Tokens meter input/output, including converted image usage.
Old Nano Banana 2 Standard output was $60 per million tokens, but resolution-dependent token use can also change. Compare actual per-image resolution rates instead of asserting every image costs half.
New input costs $1.50 per million tokens and text/thinking output $7.50, separately from image output. API Free Tier is unavailable. Search-assisted generation also needs search billing checks; budget more than image count.
Use a matching interface and response guide
The current image guide uses Interactions with gemini-nano-banana-2.1. REST is https://generativelanguage.googleapis.com/v1beta/interactions and Python client.interactions.create. Check request and response from the same guide version.
First choose the model in separate development configuration, provide input as documented and confirm SDK interface support. Do not mix another API's parameters or overwrite production before reviewing outputs.
Second inspect returned image data and dimensions before wiring storage/display. No paid calls, account access, quotas or program compatibility were tested here.
Continuous edits and wide images have conditions
Use previous_interaction_id to reference the prior interaction.id as shown. This retains context, not guaranteed unchanged people. Keep edits explicit and check people, products and text each round.
Default is 1K with 2K/4K available. Ratios 1:4, 4:1, 1:8 and 8:1 fit banners. Google says it fixed stitching artifacts at 2K/4K for these ratios; still verify composition, crop and final dimensions.
Find old 512-pixel/0.5K settings during migration: 2.1 does not support them. Select a supported size before assessing your downscaling and costs rather than copying old parameters.
The old model has no confirmed shutdown date yet
Current English notes/lifecycle docs deprecate gemini-3.1-flash-image and recommend 2.1, without a shutdown date. Deprecation calls for preparation, not proof the old API stopped.
Inventory ID, endpoint, sizes, iterative editing and price configuration, compare relevant samples, then switch gradually when accepted with configuration fallback. Continued old-model use depends on subsequent official shutdown notices.
Do not apply first-generation Nano Banana retirement to Nano Banana 2 or replace official dates with snippets. Similar names have different lifecycles; use an announced actual deadline when available.
Sources
Frequently Asked Questions
Do generated images have an AI watermark?
The guide says they include SynthID. That signal verifies neither product, text nor depicted facts.
Can a prompt guarantee an exact image count?
Docs warn image models may not follow requested counts precisely. Check actual returns and handle them programmatically.
How can a text-heavy poster reduce errors?
Google recommends generating/confirming the wording first, then placing it in the image. Check names, prices, dates and legal terms individually; improvement is no accuracy guarantee.