Nano Banana 2

Fifteen Ratios, Twenty Thousand Characters

Nano Banana 2 is Google’s current standard image model, and two things set it apart from everything else in this workspace. It renders fifteen aspect ratios — four of which, the 8:1 and 1:8 extremes, no other model here offers at all — and it accepts a brief of up to 20,000 characters, the longest prompt budget on the site. Put together, that means you can describe a full layout, in the exact shape it has to occupy, and get it back composed natively rather than cropped out of something else.

It comes in two tiers. Nano Banana 2 gives you control over the output — 1K, 2K, or 4K, PNG or JPG — while Nano Banana 2 Lite strips those choices out and renders a single size, which makes it the one to reach for while you are still deciding what the image should be. Both take the same fifteen ratios and the same long brief, and both appear in the Model dropdown above.

Four Canvases You Cannot Get Anywhere Else Here

Every other image model in this workspace tops out around 21:9. Both Nano Banana 2 and 2 Lite go to 8:1 and 1:8 — genuine strip and column formats, composed natively rather than cropped out of a landscape frame and hoping the subject survives. The light tier is not cut down here either — Lite gets all fifteen shapes too.

8:1

Announcement bars, email headers, leaderboard ads — anything that runs the full width of a page and is only a few centimetres tall.

4:1

Profile banners for LinkedIn and X, wide section dividers, hero strips above the fold.

1:4

Vertical side rails, skyscraper ad units, bookmarks, and tall sidebar art.

1:8

Roll-up and pull-up banner stands, door-height prints, spine and edge artwork.

The familiar eleven are here too — 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9, and Auto. On Nano Banana 2 every one of them renders at 1K, 2K, and 4K with nothing ruled out; on 2 Lite they all render at 1K, which is the only size it makes.

Generate It, Then Keep Working on It

Text to image and image to image run on the same model here, which matters more than it sounds. You can write a scene, generate it, then feed the result straight back in and ask for a change — and the subject, palette, and rendering style hold, because nothing has changed hands. A typical pass looks like this:

Write the scene and generate

Start in Text to Image with the ratio the final piece needs. Because the brief can run to 20,000 characters, describe the layout properly — what sits where, how it is lit, what the frame is doing at each edge.

Send the result back in

Download it, switch to Image to Image, and upload it. Describe only the thing that should change. Everything you do not mention stays where it was, including the render style the first pass established.

Repeat until it is right, then raise the size

Each pass is a fresh upload, so nothing accumulates artefacts from the last one. Settle the composition first, then re-run the winning brief at the resolution you actually need.

In Image to Image you can attach up to four references at once. The first sets the composition; the rest are things you can point at in the prompt by position — “the jacket from image 2”.

2 or 2 Lite?

The two tiers are the same model with different amounts of control. Ratios, prompt length, reference images, and both generation modes are identical; what changes is whether you get a say in the output.

Renders at 1K, 2K, or 4K — pick the size the deliverable needs
PNG or JPG, so you can choose lossless edges over a smaller file
The right tier once the composition is settled and the render ships
Small type and fine detail survive, because you can go to 4K

Nano Banana 2 Lite

One fixed size — the Resolution and Output format fields disappear
Nothing to configure beyond the ratio, so you can just keep generating
The right tier while you are still testing compositions
Same fifteen ratios and same 20,000-character brief as the full model

Both sit in the same Model dropdown, here and on the AI Image Generator and Image to Image, and switching carries your brief and your uploads across. One thing neither tier does is rendered text: if the image has to carry a headline, a paragraph, or a price that must be exact — especially in a non-Latin script — that is what Nano Banana Pro is for. Neither tier can give you a transparent background either; GPT Image 2.5 is the one with an alpha channel.

FAQ

What is Nano Banana 2?+

It is Google's current standard Nano Banana image model, available here through kie.ai. It runs both text to image and image to image on the same model, outputs up to 4K as PNG or JPG, and offers fifteen aspect ratios — including four extreme banner and column shapes no other model in this workspace has.

What is Nano Banana 2 Lite, and how is it different?+

Lite is the light tier of the same model. It produces one fixed size, with no resolution or output-format controls — those fields disappear from the panel when you select it. Everything else matches Nano Banana 2: the same fifteen aspect ratios, the same 20,000-character prompt budget, up to four reference images, and both text to image and image to image.

Which of the two should I start with?+

Lite, if you are still deciding what the image should be — there is nothing to configure beyond the ratio, so you can keep generating without stopping to think about output settings. Once you have the composition you want, switch the Model dropdown to Nano Banana 2 and re-run the same brief at the resolution and format the final piece needs.

How are these different from Nano Banana Pro?+

Reach and specialism. Both 2 and Lite add four aspect ratios Pro does not have — 1:4, 4:1, 1:8, and 8:1 — and take a longer brief, 20,000 characters against Pro's 10,000. Pro is the stronger choice in one specific case: when the image contains text that has to be rendered exactly, particularly long passages or non-Latin scripts.

What aspect ratios do they support?+

Fifteen, on both tiers: Auto, 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9, plus 1:4, 4:1, 1:8, and 8:1. On Nano Banana 2 every ratio renders at 1K, 2K, and 4K with no restrictions; on 2 Lite they all render at 1K, the only size it produces.

How long a prompt can I write?+

Up to 20,000 characters on both tiers — the largest prompt budget of any model in this workspace, double Nano Banana Pro's and ten times GPT Image 2.5's. You can describe a full layout rather than compressing it into keywords.

Can either one produce a transparent PNG?+

No. Like Nano Banana Pro, neither tier has a background control, so the background is filled even on Nano Banana 2 where you can pick PNG as the output format. Lite has no format control at all. For a genuine cut-out with an alpha channel, use GPT Image 2.5 with Background set to Transparent.

Can they edit an image I upload?+

Yes, both of them. Choose Image to Image, upload up to four images, and describe the change. The first image sets the composition and the others act as references you can name by position in the prompt.

Why does Nano Banana 2 default to JPG instead of PNG?+

That is the model's own default and it suits most uses — smaller files for photographic output. Switch the Output format field to PNG when you need lossless edges, such as flat colour, UI mockups, or anything with hard type. Lite has no format control, so the choice does not appear on that tier.

Can I use the images commercially?+

Yes, from either tier. Commercial use is included on every plan, the same as for every other model in the workspace.

Describe the Layout, Pick the Shape

Pick the shape the final piece has to fill, write the layout out in full, and generate. Settle it on Lite, then re-run the winning brief on Nano Banana 2 at the size you need.

View Pricing and Credits

Nano Banana 2: Shapes Nothing Else Here Can Make