NanoCloud

Google’s models under several names

The homepage describes NanoCloud as “a Gemini 3 Pro-based visual model” and also announces “NanoCloud Vision 2 powered by Nano Banana”. The API documentation is more precise and more useful. It lists the actual choices: nano-banana-fast, nano-banana, nano-banana-pro, two “VIP” variants and nano-banana-2. So “Vision 2” is a label on Google’s models, not a model NanoCloud trained. That’s fine as long as you know it. What you’re paying for is the interface, the credit bundles and the API wrapper, not a different engine.

The same site also offers pages for OpenAI’s Sora 2 and xAI’s Grok Imagine video models, which makes NanoCloud more of an aggregator than a single-model studio.

What the studio does well

The features are those of the underlying Nano Banana models, packaged neatly:

• Text to image and image to image at 1K, 2K or 4K, with a choice of aspect ratio, output format, style presets and prompt templates.
• Reference mixing: the homepage says up to 14 reference images, while the API documentation caps uploads at 8 per request, each up to 10 MB. Use the lower figure if you’re building on the API.
• Semantic editing: replace an object, change the mood or extend the canvas by describing it, while the lighting and perspective stay consistent.
• Consistent characters and branding across a series of images, which is where Nano Banana is strongest.

The site quotes 10–30 seconds per generation. Results come back as a task that the page polls until it finishes, and failed jobs refund their credits automatically.

Your images are public unless you switch it off

This is the most important thing to know before your first run. The generator’s “Public” setting is on by default, and the page says public images “will be displayed in Gallery”. The API works the same way: its isPublic parameter defaults to true. If you’re making product shots before a launch, client work or anything confidential, turn the setting off every time, or pass isPublic=false in every API call.

On data, the terms and privacy policy both say NanoCloud does not use your prompts, uploads or outputs to train models, and the terms give you full commercial rights to what you generate.

A watermark remover for Gemini images

Among its tools, NanoCloud offers a “Gemini Watermark Remover” that runs in your browser and strips the visible Gemini mark from images using a reverse alpha-blending technique. The homepage also lists “remove watermarks” among its editing features. Removing the mark on your own Gemini images may be legitimate, but that mark exists to show an image is AI-generated. Hiding it can breach the rules of platforms that require AI content to be labelled. NanoCloud’s own terms forbid “deceptive deepfakes or misleading content without clear disclosure”. Our view: don’t use it to pass off AI images as photographs, and never use watermark removal on images you don’t own.

Pricing: credits, and they expire

Every generation costs credits. At review time the API listed 5 credits for nano-banana-fast, 10 for nano-banana, 20 for nano-banana-pro, and 20, 30 or 50 for nano-banana-2 at 1K, 2K or 4K. The top VIP option cost up to 50 at 4K. The web generator showed 20 credits a run for its default setting. The pricing page offered three kinds of purchase, each shown with a crossed-out “original” price:

• One-time packs: about $9.90 for 2,900 credits valid for one month, $19.90 for 7,000 credits valid for three months, and $99.90 for 42,000 credits valid for six months.
• Monthly plans: about $9.90 for 4,250 credits a month, $19.90 for 10,000 (this tier adds API access and Pro models), and $99.90 for 60,000 with unlimited API calls.
• Annual plans are listed too. The pricing data didn’t make clear whether the figures shown are per month or per year, so check the checkout page.

Two points to note. Pack credits expire, so the cheapest pack gives you a month to use roughly 145 Pro-quality images. And the free side is small: a daily check-in adds 10 credits, which is half of one Pro image. The site doesn’t state how many trial credits a new account gets. The refund policy allows refunds on entirely unused credit packs within 7 days, and on unused subscription time within 14 days.

Arabic: the model may handle it, the site doesn’t say so

At review time the interface came in English and Chinese only. The homepage boasts accurate text in “English, Chinese, Japanese, and more” without mentioning Arabic, and nothing on the site addresses right-to-left text inside images. We didn’t generate test images, since that needs an account and credits. So we can’t say how well Arabic lettering renders on a poster or package design. If that’s what you need, run a few short Arabic headlines through the cheapest model first and check the letter joins and word order before buying a larger pack. Prompts themselves can probably be written in Arabic, but NanoCloud makes no promise about that either.

Who it’s for, and who should look elsewhere

NanoCloud suits designers, marketers and ecommerce sellers who want Nano Banana’s editing and character consistency in a simple browser interface. It also suits developers who want a single API key and credit balance instead of a Google Cloud account. The API docs are clear, and the auto-refund on failed jobs is a sensible touch.

Anyone already comfortable with Google’s own apps or developer tools gets the same models there, without a middleman’s markup. Teams doing confidential work need to be disciplined about the public-by-default setting. The site doesn’t name the company that runs it, and the terms refer only to “NanoCloud and its operators”, so businesses that need a named vendor and a data-processing agreement should ask first.