On October 6, 2026, Google released Nano Banana 2.1, an AI model for generating and editing images. It is an improved version of Nano Banana 2 (Gemini 3.1 Flash Image) and has been released as a stable, generally available version. Google says it significantly improves image quality, adherence to instructions, character consistency and text rendering while keeping the speed and cost efficiency of a Flash model.
Key improvements
According to Google, the main improvements over Nano Banana 2 are as follows.
- Better image quality and more accurate adherence to instructions
- Better consistency in how people and objects look across many rounds of editing
- Better rendering of text inside images
- The ability to draw on information from Google Search and Google Image Search when generating images
You can use up to 14 reference images, keeping the appearance of up to four people and the shapes of up to 10 objects consistent. Output resolution can be chosen from 1K (default), 2K and 4K.
Choose how much it thinks
With Nano Banana 2.1, you can choose how much the model thinks before creating an image from three levels: "minimal," "medium" and "high." The default is medium. You can switch between them depending on the job, for example using minimal for fast drafts and high-volume generation, and high to let it think carefully about text-heavy images and complex compositions.
In the model documentation Google published, in human preference evaluations the overall score with thinking turned on was 1050, above Nano Banana 2's 990. Google also says it improved significantly in multi-person consistency and in the factual accuracy of infographics. All of these are Google's own evaluations.
Where it is available
Nano Banana 2.1 is available in the Gemini app, AI Mode in Google Search, Google Ads, the video creation tool Flow and the design tool Stitch. For developers, it is offered in Google AI Studio and the Gemini API.
API pricing
Pricing per image in the Gemini API is as follows.
- 1K: about $0.034 (about $0.067 for Nano Banana 2)
- 2K: about $0.050 (about $0.101 for Nano Banana 2)
- 4K: about $0.113 (about $0.151 for Nano Banana 2)
For 1K and 2K images, the price is now roughly half that of Nano Banana 2. On the other hand, input pricing is $1.50 per million tokens, higher than Nano Banana 2's $0.50. If you pass in many reference images, you need to check input costs as well. There is no free tier for the API.
Limitations
In its documentation, Google also lists weaknesses such as the following.
- Small text tends to be blurry (especially in 1K images), and long passages of text may not render well
- A person's appearance may not be fully consistent between the input image and the output image
- It may get the spatial relationships between objects wrong
CoAI's take
Nano Banana 2.1 is an update that improves image quality while also lowering the price per output image. In particular, better text rendering and multi-person consistency matter in work settings such as ad banners, comics and diagrams for documents. In its developer documentation, Google positions Nano Banana 2.1 as the successor to Nano Banana 2 (no end date for Nano Banana 2 has been announced). Developers should keep in mind that while the price per image has gone down, input pricing has gone up, and estimate costs for their own usage.
