量子位官网 · 10/10/2026, 16:52:16
Google Releases Nano Banana 2.1: Enables 4K Generation and Mask Editing, Ranks Fourth in Arena Text-to-Image
Google has officially launched Nano Banana 2.1, an upgraded image generation model featuring enhanced visual design, mask editing, and subject consistency with native 4K output. The model ranks fourth in text-to-image on the Arena leaderboard, showing a ~61.7 point improvement over its predecessor, while API costs for 1K/2K images have been halved.
SOURCE COVERAGEFull Original Report
Contents4 sections
Wen Ting, reporting from Ao Fei Si
QbitAI | Official Account: QbitAI
The late-night wait was worth it—Google’s "Big Banana" has finally updated!
In the early hours of the morning, Google suddenly released its next-generation image generation model, Nano Banana 2.1, delivering comprehensive upgrades in visual design, mask editing, subject consistency, and photorealistic naturalness.
It not only generates 4K photography-grade masterpieces directly but also allows for free retouching and one-click background replacement, precisely addressing designers’ frustration with endless revisions:
△**Circle to preserve the subject and seamlessly replace the background
△**Playing chess, source: Twitter @Fries
Its performance metrics are equally impressive.
According to Arena, a leading US-based AI evaluation platform, Nano Banana 2.1 achieved an average score improvement of approximately 61.7 points over the previous generation, Nano Banana 2, across three key functions: text-to-image, multi-image editing, and single-image editing.
Specifically, text-to-image ranked fourth, showing an 80-point increase over the previous generation, second only to GPT;
Multi-image editing ranked fifth, scoring 1328 points, a 67-point increase;
Single-image editing ranked sixth, scoring 1428 points, a 38-point increase.
Even more appealing is the API pricing.
Google slashed the price for generating 1K and 2K images with Nano Banana 2.1 by half, while 4K generation costs dropped by approximately 25%.
△**AI-generated
Users couldn’t help but comment that Google’s banana is getting tastier and tastier (doge emoji).
4K Output Rivals DSLR Cameras, Improved Chinese Rendering, Max Quality
First, let’s follow netizen Christopher Fryant’s “macro camera” to experience the charm of Nano Banana 2.1’s direct 4K output:
In Christopher Fryant’s work, the man’s amber pupils feature natural color gradients and curvature, with the curl and whitening at the tips of his eyelashes clearly discernible;
Light and shadow are no longer just flat filters but carve out uneven skin textures and the slight furrow between the brows when frowning:
Even the rough texture of fabric threads is reproduced with exquisite detail:
The “photographer” himself, Christopher Fryant, couldn’t help but exclaim:
It looks exactly like the quality of photos from my DSLR camera—the cinematic feel is incredible!
Onlookers were equally amazed:
Truly astonishing, oh my god!
On the other hand, a real-world test case of a Chinese promotional poster by designer and netizen @Guizang is also well worth checking out.
After testing, Guizang’s biggest takeaway was:
The clarity of Chinese text and the issue of typos have improved significantly! There are no messy overlays or the "fragmented" look often seen in GPT outputs. The images are clean, the layout is aesthetically pleasing, the lotus flowers are elegant and beautiful, and it can even generate 4K images with one click.
It’s actually useful!
With enhanced Chinese rendering combined with price reductions, this likely means that designers in China can now use it to handle commercial projects too!
Additionally, some users have pushed the boundaries further.
Heather Cooper took images generated by Nano Banana 2.1 and fed them into Veo 3.1 for further production, ultimately creating a vivid, three-dimensional high-definition short film within Google Flow:
This means that from generating key frames and maintaining subject style consistency to bringing the visuals to life, an entire workflow can now be completed within a single platform.
Wow, designers and video editors might want to keep their Adobe Creative Suite close—those tools may soon feel threatened. (Photoshop and Premiere Pro: Please don’t come near me!)
However, before rushing to uninstall your current tools, note that Nano Banana 2.1 still has room for improvement.
Fast Speed, but Completion Quality and Stability in Continuous Editing Need Work
With Google’s release of Nano Banana 2.1, people couldn’t help but wonder: How does Nano Banana 2.1 compare to GPT Image 2.5 in terms of results?
We had to put both through rigorous real-world tests.
Netizen Zane compared the effects of ChatGPT Images 2.5 and Nano Banana 2.1 when generating an image of "Gemini-chan" on a train station platform, concluding:
Nano Banana 2.1 is indeed fast, but its visual texture and overall completion quality still fall short of GPT.
, offering 50 free credits daily. Enthusiasts can head over and try it out for free.
Additionally, developers can now access the paid API using the model name gemini-nano-banana-2.1 to perform a full suite of operations, including text-to-image generation, image-to-image transformation, poster creation, and 4K upscaling.
Without further ado, I’m eager to jump in and test Nano Banana 2.1 immediately!
Feel free to share your hands-on experiences in the comments section!
References: [1]https://x.com/NanoBanana/status/2107521202466046453 [2]https://x.com/arena/status/2107576893973344578 [3]https://support.google.com/flow/answer/16526234?hl=en
Like, Share, and Follow!
We welcome your thoughts in the comments!
— End —
🌟 Star Us 🌟