News

Releases & Models

CONFIRMED: Google Nano Banana 2.1 cuts campaign image costs, but requires an October migration

Google released Nano Banana 2.1 on October 6, with more consistent editing and a lower 1K image-output charge. Its predecessor enters the retirement schedule from October 29, while total costs depend on references and rework.

Ilustração editorial de uma banana dourada e cartões com um frasco de cosmético fictício em cenários coloridos, simbolizando edição de imagens de produto.

CONFIRMED

Google released Nano Banana 2.1 into general availability on October 6, 2026. The new version combines image generation and editing with a lower output charge at 1K resolution. For an agency or small Brazilian business producing campaign variants, the opportunity is to reduce the cost of testing settings and formats. There is also an operational deadline: its predecessor, Gemini 3.1 Flash Image, has already entered the retirement schedule from October 29.

The product needs to stay the same

According to official documentation, the update improves visual quality, instruction following, text inside images, and character consistency across successive edits. It accepts up to 14 reference images, with stated support for consistency across four characters and fidelity across ten objects. It also fixes artifacts in panoramic formats. These are features and results reported by Google, not a guarantee that every package, face, or written label will be accurate.

The economics look better at 1K

For the Gemini application programming interface, or API, Google’s official table charges $0.0336 for one 1K image output, compared with $0.067 for Nano Banana 2. The published charge is $0.0504 at 2K and $0.113 at 4K. The reduction should not be presented as uniform across resolutions. Text and media input costs $1.50 per million tokens, processing units, while text output and reasoning are billed separately. References, searches, and additional attempts therefore also affect the bill. The API does not offer a free tier for this model.

Less rework can matter more than the rate

A retailer could start from an approved product photograph and test seasonal campaign versions, changing the setting without intending to change the packaging. An agency could revise an asset conversationally while preserving the subject between requests. If the advertised consistency holds in that workload, it could save correction rounds. This does not demonstrate paid-media savings or better conversion. The useful measure is cost per approved asset, including human review and discarded generations.

The calendar makes integration urgent

Release notes announce the shutdown of the gemini-3.1-flash-image identifier on October 29. The deprecation page clarifies that its table records the earliest possible retirement dates. Anyone depending on that model should validate gemini-nano-banana-2.1 before this milestone, checking framing, Portuguese text, product fidelity, and behavior across successive edits. Updating the model name in a request without examining the resulting images is not a completed migration.

A better test score is not every asset

The model card, a document describing evaluations and limitations, reports improvements in Google’s internal tests. Independent coverage from The Decoder cautions that favorable results against Nano Banana Pro do not guarantee superior real-world quality. Our view is that version 2.1 deserves a trial using creative assets the business already approves, particularly at 1K. The central opportunity is not replacing a designer, but making experimentation cheaper without giving up visual control. The lower rate is concrete; lower rework still has to be demonstrated in the operation.

Sources

Google: October 6 release; Google: Nano Banana 2.1 features and limits; Google: current API pricing; Google: deprecation schedule; DeepMind: evaluations and limitations; The Decoder: independent coverage and caveats.