Google releases Nano Banana 2.1 image generation and editing model

Google made its image generation and editing model Nano Banana 2.1 available on October 6, 2026. Nano Banana is the name of Google's model family that creates images from a written description and lets you change an uploaded photo through conversation. The new version is an update to Nano Banana 2, also known as Gemini 3.1 Flash Image. According to the Gemini API release notes, the model became generally available the same day, and developers can call it as gemini-nano-banana-2.1.

According to Google DeepMind's model card, Nano Banana 2.1 is based on Gemini 3.6 Flash; it takes text and images as input and produces images and text. The Gemini API documentation describes it as a version that keeps Nano Banana 2's speed and cost efficiency while improving visual quality, prompt adherence, character consistency across multiple turns and text rendering inside images. Google positions Nano Banana Pro for heavier work and this model for fast, efficient tasks.

Improvements in the new version

According to the Gemini API documentation, the model produces more realistic images at 1K, 2K and 4K resolution. Tiling artifacts that appeared in 2K and 4K output at very wide and panoramic ratios such as 1:4, 4:1, 1:8 and 8:1 have been fixed. For images that contain text, such as posters, diagrams and infographics, the accuracy of the text and layout has improved. The model can combine up to 14 reference images in a single request, so the look of a person or a product can be kept consistent across several images.

While generating, the model can draw on Google Web Search and Google Image Search, which helps it depict real-world places and objects more accurately. Developers can choose how much the model thinks before generating from three levels: minimal, medium and high. Google DeepMind also says the model outperforms previous Nano Banana models in visual design, mask-based editing and subject consistency. Mask-based editing means only the marked area of an image is changed.

In the model card's evaluations, where people compare two images side by side, Nano Banana 2.1 ranks ahead of both Nano Banana 2 and Nano Banana Pro on overall preference, general editing, mask-based editing and multi-character consistency, according to the company. The new model also scores highest on a test that measures factual accuracy in infographics.

Availability and watermarking

According to the model card, Nano Banana 2.1 is available in the Gemini app, Google AI Studio, the Gemini API, AI Mode in Google Search, Google Ads, Google Flow and Google Stitch. According to the Gemini app's page, the model is open to Google AI Pro, Plus and Ultra subscribers, while free users continue with Nano Banana 2. On Google Cloud, the model is listed as generally available in the Gemini Enterprise Agent Platform documentation and currently runs in the global region. In the Gemini API, the previous model gemini-3.1-flash-image has been deprecated with no shutdown date announced, and Google recommends that developers migrate to the new model. Current prices are on Google's site.

According to the Gemini API documentation, every image the model generates includes an invisible SynthID watermark. The Gemini app also adds a visible watermark showing that the image was made with AI, and users can upload an image to the app and ask whether it was generated with Google AI. The Google Cloud documentation also lists support for C2PA Content Credentials for the model.

The model card lists several limitations: small text, long paragraphs and multi-page text are not always rendered well; character consistency between input and output images is not always perfect; instructions in mask-based editing may be followed only partly, the model sometimes confuses left and right, and its world knowledge, 3D reasoning and factuality can be limited. The comparisons were run by Google, and the model card does not list in detail how well text rendering works in each language, so it is worth testing on your own examples before relying on it for non-English text. For a fuller picture, follow evaluations from independent researchers as well.

Sources: Google DeepMind: Nano Banana 2.1 model card, Gemini API: Gemini Nano Banana 2.1, Gemini API: release notes, Gemini API: Nano Banana image generation, Gemini app: image generation, Google Cloud: Gemini Nano Banana 2.1.