Nano Banana Pro: Full Guide, Free Access from Russia and Prompts That Work

Nano Banana is the popular name for Google’s image models in the Gemini family. There are three, and confusion between them is the main reason people write "nano banana does not work": they try the basic one and expect Pro results. In order: versions, free limits, access from Russia, prompts, and what the model does worse than competitors.

Three Nano Bananas: which is which

• Nano Banana (Gemini 2.5 Flash Image): the first, fast, basic. Good for editing existing images: swap a background, remove an object, change clothes.
• Nano Banana Pro (Gemini 3 Pro Image, November 2025): the grown-up model: precise following of complex instructions, text inside the image in several languages including Russian, output up to 4K. This is what people mean by "the best model for images with text".
• Nano Banana 2 (Gemini 3.1 Flash Image): the current fast version, noticeably smarter than the first, cheaper and faster than Pro. It is the default in the Gemini app.

Simple rule: edits and quick variants, Nano Banana 2; the final frame with text, complex composition or for print, Pro. Where Pro loses to Seedream 5 Pro is in our separate comparison.

Free: where and how much

Free access exists in the Gemini app and in Google AI Studio. In Gemini, Nano Banana 2 works free at around twenty generations a day. Pro is not on the free tier: it needs a Google AI subscription. Google AI Plus (about 8 dollars) gives 50 images a day in Nano Banana 2 and Pro, Google AI Pro (about 20 dollars) 100, Ultra (about 250 dollars) 1000. Every version embeds an invisible SynthID watermark. It cannot be disabled, is not visible to the eye, but Google services recognise it. Not a problem for advertising; a problem for "passing it off as a photo", and rightly so.

Access from Russia

Entering directly from Russia, Gemini shows "not available in your country": the restriction sits on the app and on the API, and Google does not accept Russian cards. Two working paths.

1. VPN plus a Google account created from a foreign address plus a foreign card. Gives real subscription limits and the lowest price per image at volume. Downsides: the account may be blocked for jumps between addresses, and all of it needs maintaining.
2. Russian aggregators. Work through the official Google API, no VPN, rouble payment. Dearer per generation than inside a subscription, but none of the first path’s problems. For dozens of images a month a sensible choice; how to compare them is in the aggregator breakdown.

What the model does best

• Text in frame. Signs, packaging, Russian captions: Pro renders them legibly, without "hieroglyphs". Neither Midjourney nor most competitors do it as reliably.
• Editing by instruction. "Replace the mug with a blue one, leave the rest" and the rest is really left alone. A rarity.
• Several references. Give a product photo, a model photo and a location photo and ask for a scene. The product stays like itself.
• Understanding Russian. A Russian prompt works almost as well as English. For complex scenes we still write in English, but "remove the cable from the table" can be Russian.

What it does worse

• A "smooth" picture. By default the model pulls toward oversharpening and gloss: plastic skin, flat light, everything in focus. Fixed with the prompt, below.
• Artistry. Midjourney is still stronger on mood: atmosphere, painterliness, unexpected choices. Nano Banana is obedient, not inspired.
• Filters. Recognisable people, brands and some subjects are refused, sometimes without a clear reason.

Prompts that work

Structure: what is in frame, how it is shot, light, what must not appear. Separately, for any task where naturalness matters, we add a film line: it removes the AI gloss by which a picture is recognised at a glance.

Product shot: "A terracotta ceramic mug on a wooden table by a window, soft morning light from the left, light steam, blurred kitchen background. Shot on 35mm film, shallow depth of field, soft halation on highlights, fine organic grain, lifted blacks, low contrast, no sharpening, no HDR. No text, no people."

Editing an existing photo: "Remove the cable from the table and replace the plastic cup with a glass one. Keep lighting, shadows and everything else as is." One edit per request; the model is more precise than when asked for three at once.

Text in frame: "A 1970s-style coffee shop sign with the word «Morning» in large letters, warm lamps, evening, light film grain." Put the needed text in quotes and keep it to two or three words; longer phrases start to garble.

A scene from references: upload the product, the person and the place, then: "Build a scene: this person holds this product in this interior, the product exactly as in the reference, window light, film softness." The key phrase is "exactly as in the reference"; without it the model redraws the product its own way.

Step by step: the first image in Gemini

1. Open gemini.google.com (from a foreign address) or your aggregator and select the image model. Gemini defaults to Nano Banana 2; Pro is enabled separately on a paid plan.
2. Start with an edit rather than generation from scratch: upload your photo and ask for one change. In a minute you will see how the model reads instructions.
3. For generation from scratch describe the frame by the structure: what is in frame, how it is shot, light, what must not appear. Add the film line from the examples above.
4. Got almost what you need: do not rewrite the prompt. Say "keep everything, change only X". Nano Banana holds the previous frame well.
5. Make the final version in Pro at maximum resolution: the basic version for drafts, Pro for what goes to the client.

Five mistakes everyone makes

• Three edits in one request. The model does one well, two so-so, three break the image. One at a time.
• Long text in frame. A word or two is legible. A sentence comes with errors. Add slogans in layout.
• No line about light and optics. Then you get the flat gloss that gives AI away at a glance.
• A reference without "exactly as in the reference". The product gets redrawn loosely.
• Expecting Pro results from Nano Banana 2. The fast version is good, but for text and complex scenes it is another class.

What one image costs

Rough arithmetic for September 2026. Google AI Plus at 8 dollars gives 50 Pro images a day: pennies per frame with regular use, but it needs a foreign account and card. Aggregators for roubles usually charge from a dozen to a few dozen roubles per Pro image: dearer per piece, but no VPN and no account risk. The threshold where a subscription beats an aggregator is a few hundred images a month. Below that, take the aggregator and keep it simple.

Frequently asked questions

Is Nano Banana Pro available for free?

On the free Gemini tier, no: only Nano Banana 2 with a limit of about twenty generations a day. Pro opens with Google AI Plus. Russian aggregators sell Pro per image for roubles, sometimes with free starter tokens.

Where is the official Nano Banana site?

There is no separate site: the model lives in the Gemini app (gemini.google.com) and in Google AI Studio. Sites like "nanobanana.something" are intermediaries or fakes.

Can Nano Banana images be used in advertising?

Google’s terms allow commercial use of generated images on paid plans; SynthID stays in place. Separately check the rights to everything you uploaded as a reference: a stranger’s photo does not become yours because the model reworked it.

Which is better: Nano Banana Pro or Midjourney?

For precise tasks, product, text, edits, references: Nano Banana Pro. For mood and artistic frames: Midjourney. We use both in the studio; the Midjourney guide explains where it is stronger.

From image to video

For us Nano Banana Pro is the first step in almost every video: reference frames of the hero and the scene here, then animation in Seedance or Kling. That way the hero stays the same from scene to scene, and there is far less waste than generating video from scratch. The whole pipeline is in the AI video guide, and you can learn it on your own task at training.