Google DeepMind · Image Generation Model

Nano Banana 2.1

Google's new image generation and editing model, announced 6 October and built on Gemini 3.6 Flash, following Nano Banana 2 Lite (covered here in week 27). Google says it improves visual design, mask-based editing and keeping subjects consistent across edits. The model card lists up to 1M input tokens with 4K tokens of image output and 64K of text output, and a thinking-mode text-to-image preference Elo of 1050 against 990 for Nano Banana 2 and 935 for Nano Banana Pro, vendor-reported. Coverage puts the price at about half of Nano Banana 2. It is rolling out to the Gemini app, AI Studio, the Gemini API, Search AI Mode, Flow, Stitch and Google Ads. Google lists small text and character consistency as known weaknesses.