Nano Banana 2: Specs, Pricing and Ad Prompt Examples

Nano Banana 2 is the nickname of Gemini 3.1 Flash Image, the image model Google released on 26 February 2026. It is fast, relatively cheap and can blend up to 14 reference images into one picture. This page explains what Nano Banana 2 and its siblings Pro and Lite can do for your ads, what they cost and how to use them in Genly. Status: September 2026.

Short answer: Nano Banana 2 (Gemini 3.1 Flash Image) is Google's all-round image model: fast, up to 4K and with up to 14 reference images. Nano Banana Pro is the pricier option for the hardest jobs, and Nano Banana 2 Lite the fastest and cheapest, but 1K only. Small text and long paragraphs remain weak spots by Google's own account; in Genly an image costs 2 credits, or 1 with Lite.

What is Nano Banana 2?

Nano Banana is Google's nickname for its Gemini image models. Three current versions exist, all generally available in 2026 according to the Gemini API changelog:

• Nano Banana 2 (gemini-3.1-flash-image), announced 26 February 2026. Google's documentation calls it the "generalist workhorse model for all tasks".
• Nano Banana Pro (gemini-3-pro-image), since 20 November 2025: "the premium choice for the most complex visual tasks".
• Nano Banana 2 Lite (gemini-3.1-flash-lite-image), since 30 June 2026: the fastest and cheapest model.

These models "think" first, producing up to two interim images to test the composition, and you cannot switch that off in the API. On the LMArena text-to-image leaderboard (24 September 2026), Nano Banana 2 scores 1261, Lite 1250 and Pro 1247. OpenAI's models lead with more than 1400.

Nano Banana 2, Pro and Lite specs

Source: the Gemini API image generation docs, updated 23 September 2026. The only concrete speed figure Google gives is for Lite: images in 4 seconds.

Strengths and weaknesses for ads

Many references at once. This is the headline feature. You can combine your product, a model, a mood shot and a style example in a single image. Google does warn that character consistency is "not always perfect" (model card).

Text in images. Google promises "precision text rendering" and can translate text inside an image. It also lists its own weak spots: small text ("often blurry in 1k"), long paragraphs, spelling and fine details. Google's own measurements show that long text produces many errors across all models (DeepMind).

Languages. The docs list languages "for best performance", including English, German, French, Spanish, Italian, Japanese and Portuguese. Dutch, for example, is not on it. For other languages, keep copy short and proofread. Google's tip: have the model write the text first, then ask for the image with that text.

Logos and details. To keep a logo or face intact during an edit, describe it "in great detail" in your request, Google advises. Pro is Google's pick for "accurate brand consistency".

Lite. Fast and cheap, but in Google's words "not optimized for multiple reference inputs or multi-turn sequential editing". Great for quick concepts, less so for the final ad.

Pricing and access

Price per image in the Gemini API (pricing page):

The Batch API halves these prices. There is no free API tier for these image models.

In the Gemini app everyone creates images with Nano Banana 2, even without a subscription. Nano Banana Pro is available through "Redo with Pro" with a Google AI plan only. Without a plan you download at 1K, with one at 2K. Google also uses Nano Banana 2 in Google Ads, Flow and Search. Every image carries an invisible SynthID watermark.

Using Nano Banana 2 in Genly

In Genly you pick all three versions in My Projects or per image block in GenFlow:

• Nano Banana 2: 1K, 2K or 4K, up to 14 reference images, 2 credits per image.
• Nano Banana Pro: 2K or 4K, up to 8 reference images, 2 credits per image.
• Nano Banana 2 Lite: 1K, up to 10 reference images, 1 credit per image.

Aspect ratios are 9:16, 1:1, 16:9, 4:3 and 3:4; Lite adds 2:3 and 3:2. In My Projects you refer to a reference in your prompt with @image1, @image2 and so on. See the pricing page for what a credit costs. Want an AI persona in your ad? Take a look at AI UGC.

Prompt examples for static ads

Four examples with our demo brand Tropical Elixir, a tropical juice. Keep in-image text short and in quotes.

1. Product and person in one lifestyle shot

Vertical 9:16 lifestyle ad. Use @image1 for the bottle and keep its shape, label and colours exactly as shown. Use @image2 for the woman's outfit and hairstyle. She sits on a sunny balcony, seen from the side, holding the bottle, city rooftops softly blurred behind her. Warm late-afternoon light, natural skin texture. No text.

Give each reference one clear job. Without text you get a clean image to add your headline to later.

2. Flat lay of the whole range

Square 1:1 top-down flat lay on a pale sand background. Arrange the three bottles from @image1, @image2 and @image3 in a diagonal line, surrounded by fresh mango, pineapple and passion fruit halves and a few green leaves. Soft shadows, bright summer light. At the bottom, the headline "Pick your flavour." in a clean white sans-serif. No other text.

Several products in one shot is where Nano Banana 2 shines: up to 10 objects at high fidelity.

3. The same ad in another language

Edit @image1. Replace the headline with the exact German text "Sonne in der Flasche." Keep the font style, colour, position, the bottle and everything else exactly the same.

Google lists translating text inside images as a feature. German is on its best-performance list; for languages that are not, check every letter.

4. Explore concepts fast with Lite

Quick concept for a summer ad: a bottle of orange tropical juice in the front basket of a bicycle full of fresh fruit, parked by a sunny canal. Bright, playful colours, square 1:1. No text.

Swap only the setting (beach, picnic, pool) and use Lite at 1 credit per image to test directions. Then recreate the winner with Nano Banana 2 or Pro, using your real product photo as a reference.

Image models in Genly compared

Pick Nano Banana 2 when you want to combine many references. If in-image text matters most, look at GPT Image 2.5; for mood images from text alone, at Seedream.

Frequently asked questions