First order: 15 HD images for $9.90 · no watermark · commercial use included
Tool reviews

What Is Nano Banana? Google’s Gemini Image Model, Explained (Pro, 2, Prompts)

Nano Banana is the nickname for Google’s Gemini image models. Here is what each version (original, Pro, 2, 2 Lite, 2.1) can do, where it falls short, 17 copy-ready prompts, and when it arrives on PixBloom.

Editorial Team#Nano Banana#AI Models#Prompts
What Is Nano Banana? Google’s Gemini Image Model, Explained (Pro, 2, Prompts)

Nano Banana is the name almost everyone uses for Google’s Gemini image models, even though it began as an in-joke. The viral 3D figurine photos and “put these two people in one picture” edits of late 2025 were Nano Banana output. This guide explains what the name covers, how the versions differ, what the model is good and bad at, and gives you 17 prompts you can copy. It also tells you honestly where things stand with Nano Banana on PixBloom: coming soon, not live yet.

Everything factual below was checked against Google’s own announcements and documentation as of November 2026; the links are in the Sources section. We are not affiliated with Google.

Nano Banana: the nickname and the official names

In early August 2025 an anonymous image model appeared on LMArena, the public blind-comparison site, under the codename “nano-banana” and quickly became a favorite for photo editing. Google later explained that a product manager picked the name for a last-minute arena submission. On 26 August 2025 Google confirmed the model was theirs: the Gemini app update post was titled around the Nano Banana name, and the developer post introduced the same model as Gemini 2.5 Flash Image, “aka nano-banana.”

Google then kept the joke: Nano Banana Pro (November 2025), Nano Banana 2 (February 2026), Nano Banana 2 Lite (June 2026) and Nano Banana 2.1 (October 2026). So “Nano Banana” today can mean the family, the original 2025 model or whichever version the Gemini app is currently running. The table below maps each nickname to its technical name.

Nano Banana versions: what changed each time

VersionTechnical nameAnnouncedWhat it added
Nano BananaGemini 2.5 Flash Image26 Aug 2025Character consistency, blending several photos into one scene, natural-language local edits; one fixed size per ratio, roughly 1 megapixel
Nano Banana ProGemini 3 Pro Image20 Nov 20251K, 2K and 4K output; up to 14 input images and up to 5 people kept consistent; legible, multilingual text; Google Search grounding; “thinking” before drawing
Nano Banana 2Gemini 3.1 Flash Image26 Feb 2026Pro-level quality at Flash speed; 512 px to 4K; up to 5 characters and 14 objects per workflow; became the default in the Gemini app, AI Mode and Lens
Nano Banana 2 LiteGemini 3.1 Flash-Lite Image30 Jun 2026The fastest, lowest-cost member — images in as little as four seconds; meant to replace the original model for high-volume use
Nano Banana 2.1gemini-nano-banana-2.16 Oct 2026Google’s current “workhorse” model: 1K to 4K, up to 14 references (4 characters + 10 objects), tighter mask-based edits, image-search grounding, 14 aspect ratios

Three details matter in practice:

  • 4K is a Pro / 2 / 2.1 feature. Google’s docs list the 2025 model at one size per aspect ratio (1024×1024 for a square); 2K and 4K arrived with Nano Banana Pro and are now also offered by Nano Banana 2 and 2.1.
  • The original model is on its way out. Google’s model page marks Gemini 2.5 Flash Image as deprecated, retiring on 15 March 2027, and the developer docs call it “the legacy model of the Nano Banana series.”
  • Every image carries SynthID. All versions embed Google’s invisible SynthID watermark; in the Gemini app, free and AI Pro tiers also get a small visible sparkle, which Google removes for Ultra subscribers and AI Studio output.

What Nano Banana is good at

Across versions, four strengths explain the hype:

  1. Prompt following. Describe a scene in plain sentences — Chinese or English — and the model usually respects subject, setting, lighting and framing without tag soup.
  2. Multi-image references. Upload a product, a character sheet and a background mood photo, say what each image is for, and the model composes them. Pro accepts up to 14 inputs; 2.1 tracks up to 4 characters and 10 objects.
  3. Character and product consistency. The same person, mascot or product keeps its look across scenes, which is why comic series, picture books and product catalogs took off on it.
  4. Legible text. Google launched Pro as “the best model for creating images with correctly rendered and legible text,” including non-English text, and 2 / 2.1 inherited it. Short headlines on menus, posters and infographics now work.

A fifth, often overlooked: world knowledge. Sitting on a Gemini language model, it understands “Dunhuang mural palette” or “a 1990s Hong Kong film poster” without a style guide.

Known limits (as of November 2026)

  • Text still has a ceiling. Google’s docs advise generating the text first, then asking for an image containing it; long paragraphs and small type remain error-prone. Keep in-image text to a headline and a line or two, and proofread.
  • Lite trades capability for speed. Nano Banana 2 Lite is documented as “not optimized for multiple reference inputs or multi-turn sequential editing.”
  • Real people from web search are off-limits. Nano Banana 2 / 2.1 can ground on Google Search but, per the docs, do not use real-world images of people from search results.
  • Faithfulness isn’t pixel-perfect. A referenced product keeps its shape, color and material, but logos, fine print and stitching can drift. Check against the real item before publishing.
  • Access varies by surface. Free Gemini app users get limited quotas of the premium model, then fall back to a lighter one; developers use the Gemini API, AI Studio or Vertex AI; third-party sites (like ours, soon) route through the API.

Nano Banana prompts you can copy

The prompts below are written for the Pro / 2 / 2.1 generation. Swap in your own subject; keep sentences, not keyword lists.

Figurines and collectibles

  1. “Turn the character in the reference photo into a 1/7-scale collectible figurine on a round acrylic base, placed on a desk next to its retail box with the character art printed on it, soft window light, shallow depth of field, photorealistic.”
  2. “A chibi vinyl toy version of the pet in the reference, glossy finish, standing on a wooden shelf among plants, 3:4, product-photo lighting.”
  3. “Three figurines of the same character from the reference in different poses lined up on a white display shelf, consistent face and outfit, museum-style lighting.”

Product photos

  1. “The product from the reference centered on pure white, even studio light, soft natural shadow under it, no props, catalog main image, 1:1.”
  2. “The same product from the reference on a light oak table beside a cup of coffee and an open notebook, morning sunlight from the left, lifestyle photo for an online store.”
  3. “Macro close-up of the product in the reference showing texture and material, 85 mm lens look, dark neutral background, one highlight edge.”
  4. “The two products from the references side by side on a marble surface with eucalyptus sprigs, top-down flat lay, gift-set look, 4:5.”

Posters with text

  1. “A retro café menu poster titled ‘Today’s Special’ at the top, listing three drinks with prices below, hand-drawn illustration style, cream paper texture, 3:4.”
  2. “Minimalist tech-conference key visual with the headline ‘Build 2027’ in bold sans-serif, abstract blue gradient shapes, lots of negative space for a subtitle, 16:9.”
  3. “Mid-Autumn Festival poster: a glowing full moon over a tiled rooftop, osmanthus branches in the foreground, the title ‘中秋团圆’ in elegant calligraphy, warm gold and deep blue, 9:16.”

Consistent character series

  1. “A four-panel comic starring the girl from the reference: ordering at a café, waiting, receiving her latte, taking a happy selfie; same hairstyle, outfit and line-art style in every panel.”
  2. “The mascot from the reference in six emotions — happy, surprised, sleepy, angry, shy, proud — arranged as a sticker sheet on white, thick outlines, consistent proportions.”
  3. “Picture-book page: the fox from the reference walking through an autumn forest, watercolor texture, same fox design as the reference, space at the bottom for two lines of text.”

Photo restyles and edits

  1. “Keep the person and pose in the reference exactly; change the outfit to a navy wool coat and the background to a snowy Tokyo street at dusk, soft bokeh.”
  2. “Convert the reference photo into a hand-painted Japanese animated-film scene — soft watercolor skies, lush greenery — keeping the composition and the two people, warm afternoon light.”
  3. “Restore and colorize this old black-and-white family photo, repair scratches, keep faces unchanged, natural 1960s color.”
  4. “Replace the plain wall behind the subject in the reference with floor-to-ceiling bookshelves, match the lighting and perspective, leave the subject untouched.”

For the general method behind these — subject, scene, light, framing, style — see our AI image prompt guide.

How to use Nano Banana on PixBloom (coming soon)

We are preparing to offer Nano Banana Pro and Nano Banana 2 online through the model API: up to four reference images per generation, five aspect ratios, output up to 4K, prompts in English or Chinese, and the same credits-per-image model as the rest of the site. The models are not switched on yet. Open the Nano Banana page and leave your email there; we will notify you the day they go live. Until then, image-to-image on our current engines already handles product-in-new-scene and consistent-subject jobs, and the prompt library has finished examples to borrow.

To be clear about what we are: an independent site giving online access to these models to people who do not want a Google account, a developer key or a Gemini subscription. We are not affiliated with, partnered with or endorsed by Google; Nano Banana is Google’s name for Google’s models.

Sources

Nano Banana

Nano Banana Pro and Nano Banana 2 online.

Try it now

Try the High-Quality AI Image Generator Now

Write one sentence and download a sharp, watermark-free HD image about 30 seconds later.