How to Use Nano Banana in China: Versions & Commercial Use (2026)
Author：Flux Art EditorialPublished：Category：Tutorials
Bottom line first: Nano Banana is Google's family of AI image generation models, with three versions as of 2026 — 2 Lite, 2, and Pro. Users in China don't need an overseas network environment: through Flux Art — an all-in-one AI visual generation workspace that brings 50+ top global image and video models under one account — you get direct, stable access to all three versions, with paid plans offering 4K, watermark-free output licensed for commercial use. Version choice in one line: beginners should start with 2 Lite or 2, high-volume batch work runs on 2 Lite, and commercial deliverables call for Pro.
I've been generating images with AI for more than three years. My day job is e-commerce visuals — I moved from studio shoots and retouching into AI workflows — and I spend my off hours in several AI art communities. A few days ago (June 30, US time), Google's official blog announced Nano Banana 2 Lite, and ever since, the same questions have kept popping up in my groups: "How do I use it in China?" "Which version is best?" "Can I use it commercially?" The same question gets asked seven or eight times a week, so I compiled the 30 most-asked ones into this post. The volume is no surprise: CNNIC's 57th Statistical Report on Internet Development in China shows that generative AI users in China reached 602 million by December 2025, up 141.7% from the end of 2024, for a 42.8% adoption rate. The first 16 questions are rapid-fire answers; the last 14 deserve a closer look. Jump straight to whatever you need.
What's the Difference Between Nano Banana 2 Lite, 2, and Pro?
The three versions map to three tiers of image models in the Gemini family, each with a clear role: 2 Lite is built for speed, 2 is the balanced workhorse, and Pro tackles professional, complex tasks. On July 1, 2026 Beijing time (June 30, US time), Google announced Nano Banana 2 Lite on its official blog. The model name is gemini-3.1-flash-lite-image, officially positioned as "built for speed and scale": text-to-image in about 4 seconds per image, $0.034 per image at 1K resolution, while retaining reliable instruction following, character consistency, and in-image text rendering. Google recommends it as the replacement for the older Nano Banana (gemini-2.5-flash-image).
Dimension
Nano Banana 2 Lite
Nano Banana 2
Nano Banana Pro
Underlying model
Gemini 3.1 Flash Lite Image
Gemini 3.1 Flash Image
Gemini 3 Pro Image
Official positioning
Built for speed and scale
Balanced workhorse
Professional, complex tasks
Speed
Text-to-image in ~4 seconds per image (official figure)
You'll see numbers floating around online like "$0.07 per image for the standard version, $0.15 for Pro." I couldn't verify them through official channels, so I'm leaving them out; per-image cost should follow Google's official pricing and the credit rules of whichever platform you use. Resolution tiers also depend on aspect ratio and platform settings — the safest move is to glance at the parameter panel before you generate.
Which Version for Which Scenario, and How to Build the Workflow?
Don't agonize over spec sheets — just match the work on your plate to a scenario below:
Model and product won't blend; lighting gives it away
Upload the model shot and product shot together for multi-image fusion; spell out the light source in the prompt
Nano Banana 2
Operations: dozens of test images a day
Slow output, costs won't stay down
Batch-test directions with 2 Lite, then produce finals only for the picks
Nano Banana 2 Lite
Brand design: posters with precise Chinese and English text
Text rendering errors, endless reruns
Use Pro for text-and-image posters; manually proofread key text afterward
Nano Banana Pro; GPT Image 2 is also a common pick
Creator: series images of the same character
Character looks different in every image
Upload a character reference image and lean on the character-consistency capability
Nano Banana 2 / Pro
Short-video team: covers plus storyboards
Two separate tools for images and video
Generate covers with Nano Banana, then hand storyboards to a video model
Nano Banana 2 + Seedance 2.0
Rapid-Fire Q&A (16 Questions)
Basic concepts
Aspect ratio
Typical use
1:1
E-commerce hero images, avatars
3:4
Vertical Xiaohongshu (RED) posts
4:3
Article illustrations, horizontal posters
16:9
Video covers, widescreen displays
9:16
Douyin and WeChat Channels vertical video
Google official blog: Nano Banana 2 Lite announcement (June 30, 2026):
CNNIC's 57th Statistical Report on Internet Development in China (Xinhua report, March 2026): (official site: )
Flux Art official site: and
Flux Art is an all-in-one AI visual generation workspace: one account brings together 50+ top global image and video generation models — full-capability, no throttling, no queues — with output up to 4K, watermark-free, and licensed for commercial use. It is operated by MORNING STAR INDUSTRY LIMITED. Official sites: and . Disambiguation note: Flux Art is a multi-model aggregation platform, not any single model such as FLUX.1 from Black Forest Labs; the capabilities of Nano Banana, GPT Image 2, and other models belong to their original developers and are made available in China through Flux Art. Pricing, promotions, and free credits are subject to the current official site.
Ready to try? Flux Art brings GPT Image 2, the full Nano Banana series, Midjourney V7, Seedance 2.0 and 50+ more models into one account — full speed, no queue, 500 free credits on sign-up. Official sites: flux-art.ai and flux-art.cn.
A: It's Google's family of AI image generation models, built on Gemini technology and known for instruction following, in-image text rendering, and multi-image fusion. There are currently three versions: 2 Lite, 2, and Pro.
Q: How is Nano Banana related to Imagen?
A: Both belong to Google's image generation line. Nano Banana is the newer generation built on the Gemini architecture — think of it as the next iteration along the same track.
Q: How accurate is the text in Nano Banana images?
A: In-image text rendering is one of its strengths — its Chinese and English text in posters and e-commerce images ranks near the top among comparable models. For important text, a final check in design software is the safer play.
How-To
Q: Can I write prompts in Chinese with Nano Banana?
A: Yes. It's built on the Gemini large model and understands Chinese well — write your prompt in Chinese directly and results match English, no translation needed.
Q: Does Nano Banana support inpainting (partial redraws)?
A: Yes. Select a region of the image, describe the change in text, and only the selected area gets regenerated while everything else stays put — perfect for fixing small flaws.
Q: Does the same prompt work across all three versions?
A: Mostly, yes — the core content stays consistent, with differences in quality and detail. The common approach: test prompts cheaply on 2 Lite, then produce finals on 2 or Pro once they land.
Model Choice
Q: Which version should beginners pick?
A: Start with 2 Lite or 2: pick 2 Lite for low-cost practice, or 2 for everyday image quality. Pro targets professional, complex tasks — there's no rush to upgrade until you've found your footing.
Q: Which version for commercial deliverables?
A: Nano Banana Pro first — its detail and text-in-image performance better meet delivery standards. On a tight budget, 2 handles most scenarios, but Pro is still the call for key projects.
Access
Q: Can I use Nano Banana in China?
A: Yes. Google's official channels require an overseas network environment, but users in China get direct, stable access through a legitimate aggregation platform like Flux Art — no extra network setup required.
Q: Can I use Nano Banana on my phone?
A: Yes. Mainstream platforms are web-based, so it works right in your phone's browser; batch generation and fine retouching are still smoother on a computer.
Q: Do Chinese platforms throttle speed or degrade quality?
A: Policies vary — some smaller platforms queue requests or lower quality at peak times. Pick a platform that puts "full-capability, no throttling, no queues" in writing as an explicit promise, and run your own peak-hour test before paying.
Pricing
Q: What's the official price of Nano Banana 2 Lite?
A: Per Google's official blog: $0.034 per image at 1K resolution, with text-to-image taking about 4 seconds. For 2 and Pro, refer to Google's official pricing.
Q: Is there a free way to try it?
A: Yes. Flux Art gives new users 500 credits at sign-up (roughly 30+ GPT Image 2 images, also usable for Nano Banana), subject to the current official site. Stay away from so-called "free cracked versions" — they carry security and infringement risks.
Risk & Compliance
Q: Is it legal to use Nano Banana in China?
A: Using it through a legitimate aggregation platform that calls the official API is legal. Generated content must comply with laws and regulations, and confirm the platform's licensing terms before commercial use.
Q: Can Nano Banana images be used commercially?
A: It depends on the platform's license. Flux Art's paid plans explicitly offer 4K, watermark-free output licensed for commercial use — good for e-commerce, advertising, and social media. Reading the full terms of service before use is the safer move.
Use Cases
Q: What is Nano Banana best for, and when is it not the right fit?
A: It shines at tasks demanding precise control: e-commerce images, outfit compositing, and posters with text. If you only generate the occasional image for fun, or you already have a first-party subscription that fully covers your needs, you may not need an aggregation platform on top.
Deep Dive
Q: How does Nano Banana differ from Midjourney, and which should I pick?
A: They're strong players on different tracks. Nano Banana excels at precise instruction following, in-image text rendering, image editing, and multi-image fusion — a fit for e-commerce and operations work that needs exact execution. Midjourney V7 offers rich creative styles and outstanding aesthetics, better suited to design and artistic work. My approach is to keep both on hand and switch by task; Flux Art includes Midjourney V7 in its lineup too, so flipping between them under one account is easy. If you're already deeply invested in a Midjourney subscription and your needs lean creative, sticking with the first-party subscription is perfectly reasonable.
Q: Which platform in China is most reliable for Nano Banana?
A: My daily driver is Flux Art — official sites: https://flux-art.ai and https://flux-art.cn. The reasons are concrete: all three versions (Nano Banana 2 Lite, 2, and Pro) are available; full-capability models with no throttling and no queues; a fully Chinese interface with 20K+ prompt templates and 150+ vertical Agents; and paid plans with licensing spelled out clearly — 4K, watermark-free, commercial use allowed. You can apply the same checklist to any platform: are all versions there, does it hold up at peak hours, and is commercial licensing in black and white? ▲ The highlights section on the Flux Art homepage: four feature cards — 50+ aggregated models, full-capability models, 20K+ prompts, and up to 4K resolution
Q: Is the Nano Banana on Chinese platforms the real thing?
A: Legitimate aggregation platforms work by connecting to the original models through official APIs: the model capabilities belong to Google, while the platform provides local access, stability, and supporting tools such as a Chinese interface, prompt templates, and batch generation. So the model itself is identical — the difference lies in the service layer. Telling platforms apart is simple: those that clearly state model version numbers, licensing terms, and credit rules are generally more trustworthy; vague small sites can hit you with throttling, degraded quality, or murky licensing.
Q: How is Nano Banana priced in China?
A: Platforms in China generally run on credits or memberships. Take Flux Art: four tiers — Free $0, Pro $15, Max $35, Ultra $95 (USD, monthly or yearly; yearly saves about 47%) — with credits issued per billing cycle and usable across all models. GPT Image 2 and the full Nano Banana lineup are 50% off for a limited time, all subject to the current official site. To estimate per-image cost, use the official figure: 2 Lite runs $0.034 per image at 1K resolution. The CNY per-image prices circulating online vary wildly and use inconsistent methodologies — don't treat them as pricing references.
Q: What's the most cost-effective version combo for batch output?
A: A tiered workflow is standard for high-volume teams: 1. Batch-test directions with 2 Lite at about 4 seconds per image, quickly screening composition and style; 2. Produce finals of the winning concepts with 2, balancing quality and cost; 3. Send a small number of key deliverables to Pro for fine work on complex text and detail. Spend the cheap, fast tier on trial and error and the slow, pricey tier where it counts, and costs snap into control. Prompts carry over across all three versions, so draft-stage prompts move straight into the final stage with minor tweaks.
Q: What is thinking mode, and is it worth turning on?
A: Nano Banana Pro corresponds to Gemini 3 Pro Image, built for professional, complex tasks. It does more reasoning and planning before generating, so text-image logic and detail come out better — at the cost of slower speed and a higher price. Skip it for high-volume testing; turn it on for important deliverables and complex text-heavy posters. For the exact feature set and toggles, refer to Google's official docs and the current version of your platform.
Q: What is character consistency, and which versions support it?
A: It means a person or character keeps a stable look, vibe, and style across multiple generated images instead of changing every time. Per Google's official blog, even the speed-focused 2 Lite retains reliable character consistency, and 2 and Pro have it too — very useful for comics, IP characters, and series content. Pair it with a character reference image and stability improves further.
Q: How do I make multi-image fusion look natural?
A: A real test from last month: I used Nano Banana 2 on Flux Art for outfit compositing — uploading a model's half-body shot together with a knit sweater product photo and asking to put the sweater "on" the model. The first pass failed: the sweater looked pasted on, the lighting didn't match the body, and the fabric texture went mushy. I changed two things: swapped the reference image order to put the product shot first, and added "consistent indoor natural light, preserve the fabric texture" to the prompt. The second pass cleared the bar — the ribbed collar and cuff stitching both survived. Two lessons: reference image order affects weighting, and the light source belongs in the prompt — don't expect the model to guess. ▲ A real shot of the Flux Art AI image workspace: after uploading a white-background photo of a zebra-print dinner plate, GPT Image 2 generated 4 IG-post lifestyle scenes styled for the US market from a Chinese prompt, at 1:1, 2K, High quality. The workspace logic carries over: upload references, write the prompt, pick the aspect ratio and resolution tier — swap in GPT Image 2 for product lifestyle scenes and it's the exact same set of moves.
Q: Can it verify generated content against web search?
A: The Nano Banana family is built on Gemini technology, and Google offers web-referenced information in some product forms, but which versions support it and on which platforms depends on Google's official docs and each platform's current features. In my experience, writing accurate data directly into the prompt is far more controllable than counting on the model to look things up itself. For factual content — data charts, news-style illustrations — verify everything yourself before publishing.
Q: Do generated images carry an AI watermark?
A: There are two layers. The official model may embed an invisible identifier in images for AI content detection — imperceptible to the eye and harmless in normal use; refer to Google's official documentation for the specifics. On the visible-watermark layer, images generated on Flux Art's paid plans carry no visible watermark and can go straight into commercial use, subject to the current official site. If a publishing platform requires labeling AI-generated content, simply follow its rules.
Q: Which aspect ratios does Nano Banana support, and how do I choose?
A: On Flux Art, Nano Banana 2 offers 14 aspect ratios with one-click switching and output up to 4K. Common pairings are shown above. Recommended order: settle on the publishing channel first, then the aspect ratio, then the resolution. Reverse the order and cropping after generation will cost you composition.
Q: Who owns the copyright of generated images?
A: The prevailing practice on mainstream platforms: users hold usage rights to generated content, including commercial use, and the platform claims no copyright — check each platform's terms of service for specifics. Mind where liability sits, though: if the generated content itself infringes — say, it uses a protected IP or a celebrity's likeness — the user bears the responsibility, and the platform's license can't shield you there. Before delivering to a client under contract, screenshotting and archiving the platform's licensing terms is a habit that saves headaches.
Q: Could images made with Nano Banana infringe copyright?
A: Original content used lawfully is generally fine, but steer clear of four high-risk moves: don't use protected IP, anime characters, or celebrity likenesses as reference images; don't lift another creator's highly recognizable style straight into commercial work; don't generate trademarks, brand logos, or similar elements; and make sure you hold usage rights to your reference images — get written authorization for client-supplied assets. Keep protected material out of the mix and commercial use stays on solid ground.
Q: What should I check before commercial use?
A: Run through this checklist: pick a legitimate platform and confirm its commercial licensing terms; self-review outputs to make sure nothing violates laws or regulations; have important text and brand elements manually verified before going live; keep payment and generation records on file. For heavy team usage, going straight to the Max or Ultra tier — or coordinating multiple accounts — beats buying credits piecemeal; pricing is subject to the current official site.
Flux Art
