Nano Banana
Google's family of image models inside Gemini and AI Studio: Nano Banana 2 is the default, Nano Banana Pro the high-end option, and what shluchim use them for.
On this page
Nano Banana is Google's name for the image models built into Gemini. When you ask the Gemini app for a picture, a flyer background or an infographic, Nano Banana makes it. The same models run in Google AI Studio, in Google Search (AI Mode and Lens), and behind the slide decks and infographics in NotebookLM.
The models, October 2026
| Name | Technical name | Released | Notes |
|---|---|---|---|
| Nano Banana | Gemini 2.5 Flash Image | August 2025 | The original. Superseded by the newer models. |
| Nano Banana Pro | Gemini 3 Pro Image | November 2025 | Highest fidelity and the most control. Google AI Pro and Ultra subscribers can still pick it in the Gemini app. |
| Nano Banana 2 | Gemini 3.1 Flash Image | 2026-02-26 | Pro-level features at Flash speed. The default in the Gemini app's Fast, Thinking and Pro modes since launch. |
| Nano Banana 2 Lite | Gemini 3.1 Flash-Lite Image | 2026-06-30 | Fastest and cheapest; aimed at developers making many images. |
Sources: Google's Nano Banana 2 announcement, its Nano Banana Pro developer post, its Nano Banana 2 Lite post, and the Gemini Image model page.
How to get Pro in the Gemini app: Nano Banana 2 answers by default. If you pay for Google AI Pro or Ultra, generate the image, then open the three-dot menu and regenerate it with Nano Banana Pro.
Google's announcement lists what Nano Banana 2 does better than the original: it draws on web search to get real places and objects right, renders text inside images more reliably and can translate it, keeps up to five characters and 14 objects consistent across images, and outputs from 512 pixels up to 4K.
Watermarks
Every Gemini image carries an invisible SynthID mark that identifies it as AI-made. The visible mark in the corner became optional on 2026-08-14: on the web, open Gemini's settings (the cog, lower left) and choose Media watermark (TechRadar). The invisible mark stays either way. Earlier claims in the group that downloading removes the watermark were corrected the next day (November 2025).
What shluchim use it for
- Flyer backgrounds and whole flyers. From its launch in August 2025 the group treated it as the default for flyer art. Early reactions: "by far the best at understanding my requests"; the admin said its "accuracy, text and real world understanding are unmatched" (November 2025).
- Hebrew letters on flyers. In December 2025 a member called it "by far" the best for Hebrew letters. Proofread every letter anyway (see Flyer Design).
- Renovation and building renders. Upload a photo of a real room or building and ask for "more sunlight, carpet instead of flooring." A real photo makes the result much more accurate (November 2025). Members also turned a sketch of an Aron Kodesh into a realistic wood render and a prototype menorah in pieces into an assembled one (December 2025 to February 2026).
- Infographics. Upload an article, ask Gemini to summarize it, then ask it to "create an infographic" (the admin, November 2025). Nano Banana 2's web-search grounding helps here. Check every fact on an infographic before sharing it.
- Kids' activities. Children dictating coloring pages (November 2025).
Where it falls short
- Quality drops with each edit. "When you tell it to make changes, it does so perfectly, but loses image quality" (December 2025). In January 2026 the admin switched his recommendation for ads to ChatGPT's image model for this reason. A member's workaround: remove the background before editing, then put it back.
- Chabad objects. It draws the menorah with rounded or diagonal arms and the wrong number of branches, and the Luchos with rounded tops. In April 2026 a member noted Google "does not yet know to make the Menorah with straight branches or the Luchos square." In May 2026 one member had better luck with the phrase "straight armed Rambam Chabad menorah." In February 2026 Nano Banana Pro could not be talked out of putting four matzos on a seder plate. See AI Flyer Aesthetics.
- Real photos get altered. Faces and photos of the Rebbe come back subtly changed (August 2025). Place real photos yourself in Canva.
- Maps come out general, not exact. A May 2026 flyer map of a city in Gemini did not match the real streets. Give it an image of the real map.
- Gems that only write prompts. In June 2026 one member's Gemini Gems stopped making images and only returned a prompt; another said they still make images "you just need to ask for it."
Nano Banana or ChatGPT?
The group has gone back and forth. Nano Banana led from August 2025. In January 2026 the admin said the new ChatGPT image model was "better than Gemini's Nano Banana Pro" for ads and edits. In July 2026 a shliach saw a regression in ChatGPT flyers and returned to Gemini. Since September 2026 ChatGPT has Images 2.5. There is no settled winner, so the practical advice holds: run the same prompt in both and keep the better result. See Image Generation.
The "let ChatGPT write the prompt" workflow
An early tip from a shliach (August 2025) that is still in use:
- Tell ChatGPT what you want to make.
- Ask it to write a detailed image prompt with the style, layout and brand colors. The Designer Prompt does this step well.
- Paste that prompt into Gemini or Google AI Studio.
Related
- Gemini and Google AI Studio, where it lives.
- Image Generation, how it compares with the other image models.
- Canva, for making the result editable.
- Flyer Design and AI Flyer Aesthetics.
- Find it in the Toolbox.