Eight samples Google published with the launch, each shown with the prompt that produced it. Open any image to inspect the detail, then send the prompt to the workspace.
Third-party examples from Google's launch material. None of these was generated on this site.
High-quality flat lay photography creating a DIY infographic that simply explains how the water cycle works, arranged on a clean, light gray textured background. The visual story flows from left to right in clear steps. Simple, clean black arrows are hand-drawn onto the background to guide the viewer's eye. The overall mood is educational, modern, and easy to understand. The image is shot from a top-down, bird's-eye view with soft, even lighting that minimizes shadows and keeps the focus on the process.
Triptych infographic comparing three types of clouds: Cumulus, Stratus, and Cirrus. Each panel shows the cloud type in a dramatic sky with a bold label. High-contrast comic style. AR: 16:9
Create an image of Museum Clos Lucé. In the style of bright colored Synthetic Cubism. No text. Your plan is to first search for visual references, and generate after. Aspect ratio 16:9
An intimate, naturalistic cinematic close-up reveals a small, intricately illustrated sign made of recycled material, showing drawings of local birds and flowers. Delicate script below reads: "Native Wildlife: Please Observe from a Distance." Soft, diffused light filters through the leaves of a nearby fern, casting gentle shadows. The background is a soft blur of vibrant green foliage, emphasizing respect for the delicate ecosystem. Then take this concept and localize it to an Indian setting, including translation of all the text to Hindi.
Create an image of these 14 characters and items having fun at the farm. The overall atmosphere is fun, silly and joyful. It is strictly important to keep identity consistent of all the 14 characters and items.
Create a funny 6 part story with these 3 fluffy friends building a tree house. The story is thrilling throughout with emotional highs and lows and is ending in a happy moment. Keep the attire and identity consistent of all 3 characters, but their expressions and angles should vary throughout all 6 images. Make sure to only have one of each character in each image. Generate 6 images one at a time. Each image should be a separate output in 16:9 format.
This aerial shot captures a dramatic, misty landscape, likely a valley or glen, characterized by rolling, verdant hills and a winding river or loch. The photography style leans towards a moody and atmospheric aesthetic, emphasizing the grandeur and isolation of nature. The camera angle is high, looking down into the valley, providing a sweeping panoramic view that highlights the immense scale of the surroundings. The dominant colors are various shades of deep green, ranging from lush emerald in the foreground fields to darker, more muted tones on the distant mountains.
Cinematic still, evoking a vibrant, dreamlike quality often found in highly stylized musical dramas or whimsical comedies. The camera is positioned slightly low, looking up at the subject, emphasizing their commanding presence and the dramatic flair of their outfit. The color palette is exceptionally bold and high-contrast, dominated by electric blue and shocking pink, with a bright yellow accent. The background is a solid, uniform cerulean blue. The subject is a young, dark-skinned individual with tightly coiled hair, wearing an incredibly striking suit with swirling electric blue lines and hot pink concentric circles, bright yellow heart-shaped sunglasses and large pink circular earrings.
Transcript of the prompt Google publishes with this sample.
What Nano Banana 2 is
Nano Banana 2 is Google's image generation and editing model, released on 26 February 2026. Google describes it as combining the advanced features of Nano Banana Pro with the speed of Gemini Flash, and publishes it under the architecture name Gemini 3.1 Flash Image.
On this site it runs in the same image workspace as every other model. This page supplies the model, the controls this workspace exposes, published prompts you can copy, and a live quote before a task is submitted.
The distinction that matters when you choose a model is the one Google draws itself. Nano Banana Pro is positioned for high-fidelity work that needs maximum factual accuracy. Nano Banana 2 is positioned for rapid generation, precise instruction following and grounding in image search results.
Seven jobs Nano Banana 2 is suited to
The list follows what Google demonstrates with this model rather than what any reseller claims about it.
Infographics from a written brief. Google's launch samples turn a paragraph of instructions into a step-by-step flat-lay diagram and into a three-panel comparison chart with labelled panels.
Diagrams and notes redrawn as visuals. Google cites turning notes into diagrams and generating data visualisations as a consequence of the model's world knowledge.
Marketing mock-ups with legible lettering. Posters, invitations and cards work better when you name a short headline in quotation marks and say where it sits.
Localising an existing design. The model can translate the text inside an image, so an English sign can be republished for another market with the artwork intact.
Product photography and variations. Describe the change in plain language instead of drawing a mask, and state what must stay.
Storyboards with a consistent cast. Google publishes a working limit of five characters and fourteen objects held consistent in a single workflow.
Photoreal scenery and multi-format delivery. Google highlights vibrant lighting, richer textures and sharper details at Flash speed, with aspect ratios and resolutions chosen per placement.
Prompts you can copy
The four prompts below are the ones Google publishes alongside the launch samples shown in the gallery above, so each has a visible result on this page. Treat them as starting points: keep the structure and replace the subject. Opening one loads it into the workspace.
Infographic from a written brief
High-quality flat lay photography creating a DIY infographic that simply explains how the water cycle works, arranged on a clean, light gray textured background. The visual story flows from left to right in clear steps. Simple, clean black arrows are hand-drawn onto the background to guide the viewer's eye. The overall mood is educational, modern, and easy to understand. The image is shot from a top-down, bird's-eye view with soft, even lighting that minimizes shadows and keeps the focus on the process.
Three-panel comparison chart
Triptych infographic comparing three types of clouds: Cumulus, Stratus, and Cirrus. Each panel shows the cloud type in a dramatic sky with a bold label. High-contrast comic style. AR: 16:9
Readable text, then translated
An intimate, naturalistic cinematic close-up reveals a small, intricately illustrated sign made of recycled material, showing drawings of local birds and flowers. Delicate script below reads: "Native Wildlife: Please Observe from a Distance." Soft, diffused light filters through the leaves of a nearby fern, casting gentle shadows. The background is a soft blur of vibrant green foliage, emphasizing respect for the delicate ecosystem. Then take this concept and localize it to an Indian setting, including translation of all the text to Hindi.
One cast across a story sequence
Create a funny 6 part story with these 3 fluffy friends building a tree house. The story is thrilling throughout with emotional highs and lows and is ending in a happy moment. Keep the attire and identity consistent of all 3 characters, but their expressions and angles should vary throughout all 6 images. Make sure to only have one of each character in each image. Generate 6 images one at a time. Each image should be a separate output in 16:9 format.
Specifications Google publishes
These figures come from Google's own launch material for this model, not from a reseller's summary. Previews can be revised, so treat the model ID as a dated fact.
Item
Google's published figure
Architecture name
Gemini 3.1 Flash Image
Model ID in the cited post
gemini-3.1-flash-image-preview
Released
26 February 2026
Output resolutions
512 px through 4K, chosen per placement
Aspect ratios
Multiple ratios, controlled per asset
Character consistency
Up to five characters per workflow
Object consistency
Up to fourteen objects per workflow
Text
Precision rendering, plus translation of text inside the image
Grounding
Real-time information and images from web search
Provenance
SynthID watermark, with C2PA Content Credentials
Shipped in
Gemini app, Search AI Mode and Lens, AI Studio, Gemini API, Vertex AI, Flow, Google Ads
What this workspace exposes
The picker is the authority on what you can submit here. Another provider's page describes that provider's account, not this one, and price is resolved by the service rather than by this page.
For scale, fal lists 0.08 USD per image on its own model page, with 2K billed at 1.5 times the 1K rate, 4K at twice the rate, and optional web search and extended thinking billed on top. That is fal's card for fal's account and is quoted here only to show how the tier maths usually works. Your own number is the one the workspace prints.
Control
Options in this workspace
Resolution tier
1K (default), 2K, 4K
Aspect ratio
15 values, including auto, 1:1, 16:9, 9:16, 21:9, and the tall and wide extremes 1:4, 4:1, 1:8, 8:1
Reference images
Up to 14 in one request
Cost
Resolved server-side; the quote shown before submission is the binding figure
Watermarking
Not removable here; Google applies SynthID to outputs
Nano Banana 2, Pro and 2 Lite compared
All three tiers are listed on this site. The positioning row is Google's; the control rows are this workspace's own picker, which is why they can differ from what another site offers.
Because Nano Banana 2 and Pro expose the same resolution tiers and the same reference-image limit in this workspace, the difference you actually feel between them is speed and factual accuracy rather than a ceiling on output size. Run the same brief through both when the choice matters.
Fastest and most cost-efficient image model in the family, built for throughput
High-fidelity tasks that need maximum factual accuracy
Announced
February 2026
June 2026
November 2025
Model ID or architecture
gemini-3.1-flash-image-preview
gemini-3.1-flash-lite-image
Gemini 3 Pro Image, the name fal attaches to Pro Edit
Resolution here
1K, 2K, 4K
1K only
1K, 2K, 4K
Aspect ratios here
15 values, including auto and the 1:4 to 8:1 extremes
7 fixed values, no auto
11 values, including auto
Reference images here
Up to 14
Up to 5
Up to 14
Pick it when
You want fast iteration with groundable detail and legible text
You need many cheap 1K frames for ideation
You need the highest factual accuracy on a demanding brief
Editing with references instead of masks
An edit starts with images you are allowed to use and an instruction that says what should change and what must survive it. There is no mask to draw: the model reasons about which parts the instruction refers to and leaves the rest alone.
fal describes the same behaviour on its model page as semantic editing on the Flash architecture, and allows up to 14 reference images in one request for compositing, style matching or subject transfer. This workspace offers the same limit for Nano Banana 2.
Write the instruction in two halves. First the object and the change. Then the parts that must not move, such as the silhouette, the camera angle or the colour of everything else.
Check faces and hands first. Resemblance drift is the most common unwanted change in a portrait edit.
Re-read small text. Lettering, prices and legal wording are the second most common casualty.
Compare reflections, shadows and contact points where an object was added or removed.
Re-run with one instruction changed at a time, so you can attribute the difference to the instruction and not to the model's variance.
How to generate or edit here
Four steps, in the order the workspace asks for them.
Paste your prompt, or open a prompt from the gallery above with the copy button.
Choose a resolution tier and an aspect ratio. Add reference images when the job is an edit rather than a new image.
Read the quote, then sign in when the workspace asks for it.
Submit, review the result, and download the version you want to keep. If a task fails, read its status before retrying so you do not submit a duplicate.
Writing a prompt that survives revision
Google's own guidance is a formula rather than a keyword list. Name the deliverable, then the subject, the action and the scene, and build from there.
Be specific rather than short. Google's example of a weak instruction is a woman in a red dress; its stronger version adds that she is running through a park. The same page recommends deciding composition, style, image quality and aspect ratio before you generate, and gives a one-line example: generate an image of a blurry poky porcupine flying in space in the style of an oil painting with 2:3 aspect ratio.
When you revise, change one decision at a time. A prompt that edits four things at once produces a result you cannot attribute.
Limits and what this page does not promise
A model page should say where its own claims stop. These are the boundaries for this one.
No free allowance is promised. Any credits, signup grant or trial is shown in your account, not on this page.
Cost and available settings come from the live service. Figures quoted from another provider describe that provider.
Outputs carry SynthID watermarking per Google, and Google couples it with C2PA Content Credentials. This site cannot remove either.
Consistency is a tendency, not a guarantee. Google publishes the five-character and fourteen-object working limits as capabilities, not as contracts.
Text inside an image still needs proofreading at full size before it reaches print, a client or a legal notice.
The gallery images are third-party examples. None of them was generated on this site, and the published prompts are not a guarantee that a given result will reproduce.
The model ID above is published as a preview identifier. Preview availability and behaviour can change without notice.
Where the samples and specifications come from
Every figure and every image on this page traces back to one of the sources below. Where a value comes from a reseller rather than from Google, the row says so.
Source
Used for
Reference
Google, launch post, 26 February 2026
Model identity, capabilities, consistency limits, grounding, provenance, availability
The same source photo and its Nano Banana 2 edit, as labelled in the third-party gallery these frames come from. A pixel correlation of 0.99 between the two frames confirms the composition was kept rather than replaced.
Third-party pair from banananano.ai's model-comparison gallery, where the two frames are labelled 'Original watch image' and 'Nano Banana 2 edited watch'. Neither frame was generated on this site, the prompt behind the edit was not published, and the captions repeat the source's own labels rather than an independent test. · Third-party visual reference reused at the site owner's request; no stock-image licence is granted and no endorsement is implied.
One of the two example reference images fal loads by default in its Nano Banana 2 edit playground, where the prompt field reads: make a photo of the man driving the car down the california coastline. It is shown to illustrate the reference-image workflow, not as a result produced here.
Provider example asset from fal.ai's Nano Banana 2 edit page. fal publishes no caption or prompt binding for this specific file, so it is presented as the provider's own asset rather than as a verified before-and-after, and this site has not run the model against it. · Third-party visual reference reused at the site owner's request; no stock-image licence is granted and no endorsement is implied.
Questions
What is Nano Banana 2?
It is Google's image generation and editing model, released on 26 February 2026 and published under the architecture name Gemini 3.1 Flash Image. Google's own summary is that it brings Nano Banana Pro capabilities to Gemini Flash speed.
Can I use Nano Banana 2 for free here?
This page promises no free generations and no signup allowance. Check your account balance and read the quote the workspace shows before you submit. Whatever credits and payment options exist are listed in your account.
How do I edit a photo instead of generating a new image?
Upload the reference, then describe the change and the parts that must stay. State the object and the adjustment in one sentence, and add a second sentence naming what should not move. Compare the result with the source for faces, proportions, reflections and small text.
Will the same character survive across several images?
Google publishes a working limit of five characters and fourteen objects held consistent in one workflow, which is what makes storyboards possible. Keep the reference set attached to every request and name the character each time. Resemblance is a tendency rather than a guarantee, so review each frame.
Can it write readable text and translate text inside an image?
Yes, on both counts, according to Google. Legible lettering is intended for marketing mock-ups and cards, and the model can translate the text already inside an image so an existing design can be reused in another language. Proofread anything critical at full size.
How does Nano Banana 2 differ from Nano Banana Pro?
Google splits them by purpose. Pro is for high-fidelity work that needs maximum factual accuracy, and Nano Banana 2 is for rapid generation, precise instruction following and image-search grounding. In this workspace both expose 1K, 2K and 4K and up to 14 reference images, so the practical difference is speed and accuracy rather than maximum resolution.
And what is Nano Banana 2 Lite?
Google's fastest and most cost-efficient image model in the family, announced on 30 June 2026 under the model ID gemini-3.1-flash-lite-image and built for high-throughput work. In this workspace it is fixed at 1K with five reference images and seven aspect ratios.
Which resolutions and aspect ratios can I choose?
This workspace offers 1K, 2K and 4K, with 1K as the default. The aspect ratio list has 15 values and includes auto, the common 1:1, 16:9 and 9:16 frames, 21:9, and the extreme 1:4, 4:1, 1:8 and 8:1 ratios. Other sites list a different set, and their list does not apply to this account.
Are the images in the gallery generated on this site?
No. They are third-party samples: Google's launch images with the prompts Google published, one example reference image from fal's edit playground, and two frames from banananano.ai's model-comparison gallery. Each one carries its source in the caption.
Does the output carry a watermark or a provenance marker?
Google applies SynthID watermarking to outputs and pairs it with C2PA Content Credentials. Google's own instructions say you can upload an image to the Gemini app and ask whether Google AI generated it. Nothing on this site removes either marker.
What does it cost per image?
The workspace quote is the only number that applies to your task, and it is resolved by the service before submission. For scale, fal lists 0.08 USD per image with 2K at 1.5 times and 4K at twice the base rate. That is fal's own rate card for fal's account.
How should I compare it against GPT Image 2.5 or Seedream?
Run the same brief and the same permitted reference through each workspace, then compare text accuracy, subject detail, composition, elapsed time and the quote you were shown. Choose per deliverable rather than assuming one model wins every task.
Explore details ↗ 
