Google AI image model Nano Banana 2.1 released to improve visual design, mask editing, and subject consistency谷歌 AI 图像模型 Nano Banana 2.1 发布，提升视觉设计、蒙版编辑、主体一致性

Google today officially released Nano Banana 2.1, an upgraded version of the company's AI image generation and editing model. Google emphasized that the new version of Nano Banana 2.1 has achieved all-round improvements, especially in visual design, mask-based image editing, and subject consistency. Earlier this year, Google launched Nano Banana 2, also known as Gemini 3.1 Flash Image. The Nano Banana 2 achieves the same level of graphics quality as the Nano Banana Pro, while boasting the high-speed features of the Google Flash family of models. Google has also downgraded a number of features that originally belonged to the Pro version to this lower-positioned model, including real-world knowledge base, text rendering effect optimization, and character and object consistency performance improvement. IT House notes that Nano Banana 2.1 builds on its original capabilities and improves in three main directions: Visual Design: Nano Banana 2.1 produces better composition and more complete visuals. Mask editing: Users can more accurately select and modify local areas of an existing image without having to regenerate the entire screen. Subject consistency: When performing editing operations or generating multiple associated images in batches, the model can better preserve the appearance and characteristics of the main body of the screen. The improved subject consistency is very practical for Gemini users, which can ensure that the same person or object remains unified in the appearance of multiple generated results. Google Nano Banana 2.1 is now gradually deployed, covering the Gemini app, the AI model of Google Search, Google AI Studio, Google Flow, Stitch, Google Ads, and the Gemini Enterprise Platform. For developers, Nano Banana 2.1 has been officially opened, corresponding to the model ID gemini ‑ nano ‑ banana ‑ 2.1; the model supports text, picture, audio, and video inputs, and can output images in 1K, 2K, and 4K resolutions. Today, OpenAI's ChatGPT Images 2.5 remains the world's leading image generation editing model. When the user continues to fine-tune an image, it is better at retaining the previous round of modifications, does not change areas that have not been edited, and does not deteriorate the quality of the original image with repeated iterations. The OpenAI product is followed by Microsoft AI's Mai ‑ Image ‑ 2.6 model, which was released in August this year.