Google releases Nano Banana 2.1 image model with mask-based editing gains

Google has released Nano Banana 2.1, an image model that Google says outperforms its previous models across the board in visual design, mask-based editing, subject consistency and natural-looking imagery.

Google says it can be tried now in the Gemini app and Google AI Studio, and Josh Woodward, who leads Google's Gemini app work, calls mask-based editing his favorite improvement. These claims come from Google and Woodward's own announcements; no independent evaluation is cited.

QbitAI reports that the model ranks fourth in text-to-image on Arena, 80 points above its predecessor, and that Google halved API prices for 1K and 2K images and cut 4K prices about 25%. The same report says Google gives free users 50 credits per day and that developers can access the model through the paid API as gemini-nano-banana-2.1. QbitAI's testers say generation is faster than GPT Image 2.5 but the finish is weaker, and they report a watermark bug that degrades images over repeated edits.