Nano Banana Explained: Every Version, Price and How to Use It

There isn't one Nano Banana. There are four, and Google named the first one at 2:30 in the morning.
The short answer: Nano Banana is Google DeepMind's family of image generation and editing models, built natively into the Gemini architecture. As of 19 August 2026 the family has four members: the original Nano Banana (Gemini 2.5 Flash Image), Nano Banana Pro, Nano Banana 2 and Nano Banana 2 Lite. Because these models reason over text and images together, they're unusually good at conversational editing and keeping the same character across a whole series. Most explainers online still describe a single model. That stopped being true in November 2025.
If you're choosing between tools rather than versions, start with our guide to the best AI image generator in 2026. This page untangles the bananas.
The version map: one brand, four engines
First, the name. Google product manager Naina Raisinghani coined it as a placeholder for an anonymous LMArena test in August 2025. On Google's own blog she recalls the 2:30 a.m. message asking for a submission name and her reply: "OK, how about something funny like 'Nano Banana'?" It's a mash-up of her nicknames, Naina Banana and Nano, and it stuck because the mystery model went viral before Google could rename it. On 26 August 2025 Google confirmed the model was Gemini 2.5 Flash Image, then quietly adopted the joke as an umbrella brand.
Here's the map, verified against Google's API documentation on 19 August 2026:
Nickname | Official name | Model ID | Launched | Resolutions | Status |
|---|---|---|---|---|---|
Nano Banana | Gemini 2.5 Flash Image | gemini-2.5-flash-image | Aug 2025 | 1024px | Legacy; retires on the Gemini Enterprise Agent Platform on 2 Oct 2026 |
Nano Banana Pro | Gemini 3 Pro Image | gemini-3-pro-image | 20 Nov 2025 | 1K / 2K / 4K | Current, studio tier |
Nano Banana 2 | Gemini 3.1 Flash Image | gemini-3.1-flash-image | 26 Feb 2026 | 512px / 1K / 2K / 4K | Current, app default |
Nano Banana 2 Lite | Gemini 3.1 Flash-Lite Image | gemini-3.1-flash-lite-image | 30 Jun 2026 | 1K | Current, volume tier |
Two details worth pinning. Google Cloud's own getting-started notebook gives the original model a hard retirement date on the enterprise platform, and Google's docs strongly recommend legacy users migrate to 2 Lite for better quality at lower cost. And in the Gemini app none of these names appear at all: you pick Fast, Thinking or Pro, with a "Redo with Pro" button for paid users. One brand, four engines, three consumer labels. No wonder the search results are a mess.

What each version is best at
Nano Banana (original). The August 2025 viral hit that, by Google's own telling via press, pulled tens of millions of new users into Gemini within two weeks and powered the 3D-figurine and saree trends. It's now the legacy option: fine at 1024px, cheap, and first in line to retire.
Nano Banana Pro. The reasoning-heavy studio model. Native 2K and 4K, Search grounding for factual infographics, up to 14 reference images, and the best text rendering in the family. Google's own documentation admits it still fumbles small faces and fine spelling.
Nano Banana 2. The everyday workhorse and the Gemini app's default since February 2026. Google pitched it as Pro-level quality at Flash speed, and the blind-vote arenas broadly agree (more on those below). It maintains up to 5 consistent characters and blends up to 14 images.
Nano Banana 2 Lite. The volume sprinter: 1K only, roughly 4-second generations, and the cheapest Google image model. Built for high-throughput work where good-and-fast beats perfect.
How much does Nano Banana cost?
Google bills image output in tokens, which is how one brand ends up with seven price points. Converted to per-image figures from Google's own pricing page, verified 19 August 2026:
Model | Resolution | Per image (standard) |
|---|---|---|
Nano Banana | 1024px | $0.039 |
Nano Banana Pro | 1K or 2K | $0.134 |
Nano Banana Pro | 4K | $0.24 |
Nano Banana 2 | 512px | $0.045 |
Nano Banana 2 | 1K | $0.067 |
Nano Banana 2 | 2K | $0.101 |
Nano Banana 2 | 4K | $0.151 |
Nano Banana 2 Lite | 1K | $0.034 |
Batch processing halves every figure above, so 2 Lite bottoms out around $0.017 an image.
Three things almost every pricing article gets wrong. First, 1K and 2K cost the same on Pro (both are 1,120 output tokens), so 2K is the sensible default there. Second, Search grounding is billed separately: 5,000 free prompts a month across Gemini 3, then $14 per 1,000 queries. Third, a per-image price without a model ID, a resolution and a date is meaningless. Plenty of ranking pages still quote a flat "$0.039", which only ever applied to the original model at one resolution.
If you'd rather skip the token maths entirely, the same Gemini models run inside JammyJar: one prompt box, the right model suggested for the job, no setup.

Is Nano Banana free?
Yes, with volatile small print. Free Gemini app users get around 100 images a day on the standard model, but Nano Banana Pro is capped at a hard 2 images a day. That's down from 3: Engadget reported the cut in November 2025, citing a Google support document first spotted by 9to5Google, which warned that "Limits may change frequently and will reset daily". When your free Pro quota runs out, PCMag notes the app drops you back to the standard model. No banner, no apology. A silent downgrade.
Since I/O 2026, Google has also moved the app to compute-based limits that refresh every 5 hours up to a weekly cap and factor in prompt complexity, so any fixed daily number is a snapshot, not a promise. We haven't been able to pin the exact free allowance for August 2026; right now, nobody outside Google can. Paid tiers raise the caps: per PCMag's reporting on Google's support pages, AI Pro ($19.99/mo) includes 100 image-generation credits a day, and AI Ultra subscribers can make 1,000 Nano Banana Pro images daily. AI Plus sits at $4.99 a month after a June 2026 price cut; Ultra runs $99.99 or $199.99. Even paying users report "503 Model Overloaded" errors at peak, so budget for retries. Google AI Studio, meanwhile, stays the generous route for testers: free, rate-limited, no card.
How to access it, from Gemini to Photoshop
Access is the family's real moat. The models are everywhere:
- Consumer: Gemini app, Google Search AI Mode, Google Lens, NotebookLM, Google Photos.
- Work: Google Workspace (Slides, Vids, Docs), Google Ads, the Gemini Enterprise Agent Platform.
- Builders: Google AI Studio and the Gemini API.
- Creative tools: Google Flow, Adobe Firefly, Photoshop's Generative Fill, and Figma via the Firefly plugin.
- Third-party hosts: OpenRouter, fal.ai, Replicate, Together AI, Leonardo and others, at their own prices.
One caveat hides in Adobe's fine print: Firefly carries Gemini 2.5, 3 and 3.1 as partner models but labels only Adobe's own models "Commercially Safe", which quietly hands rights due-diligence back to you. And juggling five surfaces to reach one model family is exactly the tab-chaos JammyJar exists to end; our walkthrough on how to use Nano Banana without the setup covers the two-minute version.

Nano Banana vs the alternatives
Here's the part Google's marketing won't volunteer: as of 19 August 2026, these models no longer top the blind-vote arenas they made famous. On Artificial Analysis's Text-to-Image leaderboard, OpenAI's GPT Image 2 (high) leads at Elo 1,369 across 14,483 samples; Nano Banana 2 sits third at 1,320, Pro sixth at 1,296, and 2 Lite seventh at 1,291. On the Image Editing board, Reve 2.1 leads at 1,263 while Nano Banana 2 places sixth at 1,250. The original model has slid to #41 on Text-to-Image. Google's #1 announcements from the November 2025 Pro launch, amplified by Jeff Dean at the time, are simply out of date.
Read the numbers with care, though. Arena Elo measures aesthetic preference on generic prompts, not editing fidelity or text accuracy. The editing top three sit within roughly 1% of each other, close enough that ties are broken by confidence intervals rather than raw score. And several entries still wear "Preview" tags, so positions move.
The practitioner consensus across 2026 comparisons is more useful than any single rank: GPT Image 2 wins accuracy and text, Midjourney wins aesthetics, FLUX.2 wins open-weight control, Recraft is the only major model producing true SVG, and the Nano Banana family wins value, speed, editing and character consistency. Note that last gap. No Nano Banana model outputs SVG; if you need vectors, that's Recraft's territory, and a workspace that routes each job to the right model beats forcing one family to do everything. Full rankings live in our best AI image generator 2026 guide.

Watermarks, ownership and commercial use
Every output from the family carries an invisible SynthID watermark, and newer models add interoperable C2PA Content Credentials on top; a visible Gemini sparkle appears only on free-tier images. The SynthID-Image paper on arXiv documents both the internet-scale approach and its limits: the mark degrades under aggressive edits, and detection requires Google's own infrastructure.
Commercially, you own your outputs under Google's terms and can use them in ads, client work and products. Two caveats. Ownership doesn't extend to third-party IP referenced in your prompts, and purely AI-generated works may not qualify for copyright at all under current US Copyright Office guidance, so plan brand-critical assets accordingly. Watermark-removal services exist; using them risks a terms violation plus disclosure-law exposure under the EU AI Act's Article 50 and California's SB 942. Not worth it.
FAQ
Is Nano Banana the same as Gemini 2.5 Flash Image?
Originally, yes. Nano Banana began as the LMArena codename for Gemini 2.5 Flash Image, which Google confirmed on 26 August 2025. Since then the nickname has grown into an umbrella brand covering four models: the original, Nano Banana Pro (Gemini 3 Pro Image), Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana 2 Lite.
What's the difference between Nano Banana 2 and Nano Banana Pro?
Nano Banana 2 is the fast everyday default: $0.067 per 1K image, four resolutions, strong editing and consistency. Pro is the reasoning-heavy studio model: better text rendering, Search grounding and native 4K, at $0.134 to $0.24 per image. Use 2 for volume and iteration, and Pro when typography or factual infographics matter.
Can I use the images commercially?
Yes. Under Google's terms you own your outputs and can use them commercially, including client work and advertising. You don't gain rights to third-party IP referenced in your prompts, and purely AI-generated works may not qualify for copyright under current US Copyright Office guidance, so treat brand-critical assets with extra care.
Does Nano Banana make SVGs?
No. Every model in the family outputs raster images only. Among the major generators, Recraft is the one producing true vector SVG. If your workflow mixes raster concepting with vector deliverables, use a workspace that routes each job to the model that can actually do it, rather than forcing one family to cover everything.
When does the original model retire?
The original Nano Banana (Gemini 2.5 Flash Image) retires on the Gemini Enterprise Agent Platform on 2 October 2026, per Google Cloud's own documentation, and Google recommends migrating to Nano Banana 2 Lite for lower cost and faster generation. Pro, 2 and 2 Lite all remain current as of August 2026.