Nano Banana Pro
Last reviewed
May 31, 2026
Sources
10 citations
Review status
Source-backed
Revision
v2 · 2,075 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 31, 2026
Sources
10 citations
Review status
Source-backed
Revision
v2 · 2,075 words
Add missing citations, update stale details, or suggest a clearer explanation.
Nano Banana Pro is a professional grade image generation and editing model from Google DeepMind, released on November 20, 2025. Nano Banana Pro is the consumer facing nickname for a model that Google officially calls Gemini 3 Pro Image. It builds on the Gemini 3 Pro reasoning model, and it is the higher end sibling of the original Nano Banana, the viral image tool that Google launched a few months earlier as Gemini 2.5 Flash Image. The Pro version keeps the playful name but aims at a more demanding audience, with headline features that include output up to 4K resolution, much sharper text inside images, the ability to ground a picture in live data from Google Search, and consistency across many reference images. [1][2][3]
The original Nano Banana spread quickly after its August 2025 debut. It drove a wave of viral edits, from figurine portraits to retro photo styles, and Google says the Gemini app powered more than 5 billion creations with it in the months that followed. [1][4] That popularity is part of why the Gemini app shot up the app store charts that autumn. [4] Nano Banana Pro is the answer to what those users asked for next, which was more resolution, more control, and text that actually reads correctly. It sits inside the broader family of text-to-image models and modern AI image generation systems, and it ships across Google's consumer, developer, and enterprise products at the same time.
There are two Nano Banana models, and the difference matters. The base Nano Banana runs on Gemini 2.5 Flash, a fast and cheap model tuned for quick edits and casual creation. Nano Banana Pro runs on Gemini 3 Pro, the larger reasoning model from the Gemini line. [1][3] Because it inherits that reasoning ability, Nano Banana Pro can do a short planning step before it draws anything. Google describes this as a thinking process, where the model works through composition and content first and can produce interim drafts on the way to the final image. [5]
In the Gemini app the split shows up in the model picker. The faster default uses the base Nano Banana, while choosing the thinking option routes the request to Nano Banana Pro. [6] So the base model is not a stale duplicate that the Pro replaced. The two coexist, with the base aimed at speed and volume and the Pro aimed at quality and accuracy. People who want a fast meme edit can stay on the base, and people who want a legible 4K poster reach for the Pro.
Nano Banana Pro generates images at 1K, 2K, and 4K resolution, which is a clear step up from the single lower resolution tier of the base model. [2][10] It supports a wide set of aspect ratios, from square 1:1 through portrait 9:16 and widescreen and cinematic formats, so the same model can produce a square social post, a tall phone wallpaper, or a wide banner. [2][5]
The higher resolution is the feature people notice first, because it makes the output usable for print mockups, presentation slides, and detailed scenes where small elements would smear at lower resolution. The 4K tier costs more to run, which Google reflects in both token counts and API pricing, covered in the availability section below.
Legible text has been a long standing weakness for image models, and it is one of the areas where Nano Banana Pro improved the most. The model can render longer passages of readable text directly in a picture, rather than the garbled lettering that older systems produced. [2][3] It handles multiple languages, which helps with localization, and it can style text with specific fonts, textures, and calligraphy effects. [2]
This turns the model into a practical tool for posters, greeting cards, product mockups, diagrams, and menus, where the words have to be correct and not just decorative. Several reviewers singled out text rendering as the most obvious jump over the original Nano Banana. [3][7]
The most distinctive capability is grounding an image in real information through Google Search. Nano Banana Pro can pull live, factual data and then lay it out as an infographic, a diagram, or a data visualization. [1][5] Because the model checks real sources, the numbers and labels in a chart are more likely to be accurate rather than plausible looking but wrong, which is the usual failure mode for AI generated infographics. [8]
Google frames this as connecting image creation to Gemini's world knowledge and reasoning, so a prompt can ask for something like a weather summary, a sports recap, or a recipe card and get a visual that reflects current real data. [1] Reviewers treated the more accurate infographics as the feature that sets the model apart from rivals. [8] The accuracy still depends on what Search returns, so it reduces errors rather than removing them entirely. [8]
Nano Banana Pro can take up to 14 reference images as input and blend them into a single composition. [1][2][10] It can keep up to 5 people looking like themselves across different generations, which is useful for storyboards, brand characters, or any project that reuses the same faces. [1][5] Within that budget Google says the model can hold up to 6 objects at high fidelity while drawing on the wider set of inputs. [5] That is a meaningful jump in how much visual context the model can juggle at once.
On top of that it offers studio style editing controls. Users can change the camera angle, adjust focus and depth of field, apply color grading, and transform the lighting of a scene, for example turning day into night or adding bokeh. [1][2] Edits can be localized to one part of an image instead of regenerating the whole frame. [6] These controls move the model closer to a directable production tool than a single shot generator.
Every image from Nano Banana Pro carries an invisible SynthID watermark, the digital signature Google DeepMind uses to mark AI generated content. [1][2] The Gemini app also lets people upload an image and ask whether it was made or edited by Google's AI, which checks for that SynthID signal. [1]
There is also a visible mark. Images made on the free and Pro tiers include a visible Gemini sparkle logo in the corner. Google AI Ultra subscribers, and developers using the API, can generate images without that visible logo, while the invisible SynthID watermark stays in place regardless. [6] In other words the visible badge is a tier feature, but the provenance signal is always present.
Nano Banana Pro launched across many surfaces at once. For consumers it is in the Gemini app, in NotebookLM, and in Workspace apps including Google Slides and Google Vids, and it is built into Google Ads. [1] Integration with Google Search in AI Mode was announced as coming soon for certain queries. [1] For developers it is available in the Gemini API and in Google AI Studio, and for enterprise it runs on Vertex AI and Gemini Enterprise, as well as Google's agentic development platform Antigravity. [1][5]
Access in the Gemini app follows the subscription tiers. The model is available to everyone on the free tier with usage limits, and Google AI Plus, Pro, and Ultra subscribers get progressively higher limits. [4][6] Search's AI Mode integration is gated to the paid Pro and Ultra tiers. [6] On the API the model is named gemini-3-pro-image-preview, and developers reach it through the Gemini API and Google AI Studio. [5][9] Pricing runs on tokens rather than a flat per image rate. Image output is billed at 120 dollars per 1 million tokens. A 1K or 2K image uses about 1,120 tokens and costs roughly 0.134 dollars, while a 4K image uses about 2,000 tokens and costs roughly 0.24 dollars. [8][9] That makes the highest resolution close to double the price of the standard one, so cost scales with how much detail a project needs. For comparison, a base Nano Banana image runs about 0.039 dollars, so the Pro model is several times more expensive per picture. [8]
| Attribute | Detail |
|---|---|
| Developer | Google DeepMind |
| Official model name | Gemini 3 Pro Image |
| Consumer name | Nano Banana Pro |
| Base model | Gemini 3 Pro |
| Release date | November 20, 2025 |
| API model ID | gemini-3-pro-image-preview |
| Resolutions | 1K, 2K, 4K |
| Aspect ratios | Multiple, from 1:1 square to 16:9, 9:16, and ultrawide |
| Reference images | Up to 14 |
| Person consistency | Up to 5 people |
| Text rendering | Long, legible, multilingual passages |
| Search grounding | Yes, via Google Search for factual infographics |
| Studio controls | Camera angle, focus, depth of field, color grading, relighting |
| Provenance | Invisible SynthID watermark, plus visible logo on free and Pro tiers |
| API pricing | 0.134 dollars per 1K/2K image, 0.24 dollars per 4K image |
Early coverage treated Nano Banana Pro as a serious upgrade rather than an incremental one. TechCrunch noted that the model improves on its predecessor with more detailed images and accurate text in different styles, fonts, and languages, while observing that it is slower and costlier than the original. [7] Independent reviewer Simon Willison called it the best available image generation model, and developer Max Woolf wrote that it is the best AI image generator, with caveats, singling out the text quality and the higher resolution while flagging cost and the visible watermark. [8][9]
The model lands in a crowded field. It competes most directly with OpenAI's GPT Image, the system behind image creation in ChatGPT, and with ByteDance's Seedream 4.0, which also pushed into high resolution multi reference generation in 2025. [7] Against those rivals, Google's pitch leans on the Search grounding and the Gemini reasoning underneath, which competitors do not offer in the same way. The base imagen 4 line and earlier systems like DALL-E sit in the same broad market but target different tradeoffs of speed and cost.
Nano Banana Pro is powerful but not flawless. Search grounding lowers the rate of wrong facts in infographics, yet the output is only as reliable as what Search returns, so charts can still contain inaccuracies and should be checked before anyone trusts them. [8] The 4K tier and the higher rate limits sit behind paid plans, and the free tier caps how many images a person can make. [3][6] The visible Gemini logo on free and Pro images is fixed unless a user moves to the Ultra tier or the API. [6]
The model also carries the usual caveats of AI image generation. It can still misread an unusual prompt, and the thinking step adds latency compared with the faster base Nano Banana, so it trades some speed for quality. As of launch the API model was labeled a preview, which signals that behavior and pricing could change as Google moves it toward general availability. [2]