Overnight Drop: Native 4K and Half-Priced APIs

In the early hours of October 10, Google quietly released Nano Banana 2.1, the next generation of its image-generation model, upgrading visual design, mask editing, subject consistency and overall realism: native 4K output, one-click background replacement and free-form retouching. Even more striking is pricing — generation at 1K/2K resolutions is cut in half, 4K drops by roughly 25%, and the paid API is open to developers under the model name gemini-nano-banana-2.1.

Arena Scores and the Chinese Typography Leap

Per the U.S. evaluation platform Arena, the new model gains an average of about 61.7 points across text-to-image, multi-image editing and single-image editing versus Nano Banana 2: text-to-image rises to fourth (+80 points, just behind the GPT family), multi-image editing to fifth (1328, +67) and single-image editing to sixth (1428, +38).

Nano Banana 2.1 at a glance

· Native 4K output + mask editing + subject-consistency upgrades
· 1K/2K generation prices halved; 4K down ~25%
· Live on Gemini App and Google AI Studio (built on Gemini 3.6 Flash), with 50 free credits daily
· Chains with Veo 3.1 and Google Flow for an end-to-end image-to-video workflow

For Chinese-speaking users, the more consequential change is text rendering: in designer @guizang's Chinese poster tests, images showed none of the drifting artifacts or "shattered" text common in rivals, with clean layout and one-click 4K — combined with the price cut, the cost threshold for domestic design teams to take on commercial work drops markedly. Creators can also hand generated images to Veo 3.1 and finish a keyframe-to-video pipeline inside Google Flow.

Fast, but Still Short in Spots

Generation speed is genuinely fast, but multi-round editing stability isn't ready to "replace your retouching suite" just yet.

Testers report that against GPT Image 2.5, Nano Banana 2.1 wins on speed while trailing slightly on texture and overall polish; successive editing rounds introduce noise and color drift, with a suspected watermark-stacking bug — an "edit it until it blurs" experience that hurts for professionals. On balance, Google's upgrade-plus-price-cut combo keeps compressing unit prices across the image-generation track, and the practical gains in Chinese make multimodal content production cheaper still — worth evaluating for marketing and government content pipelines at scale. NineZenith Research will keep tracking how it fares in enterprise multimodal workflows.

(Compiled from QbitAI, Google's official announcements, Arena leaderboards and hands-on coverage by BigGo and other outlets)