Nano Banana 2.1

Nano Banana 2.1 is the current fast image model in Google's Nano Banana family. It generates pictures from text, edits existing images, and combines up to fourteen references in one request. Give each reference a clear role: the product to preserve, the setting to borrow, or the composition to follow. Subject pictures and recipe anchors count toward the same fourteen-image capacity.

Its resolution ladder starts at 1K for 3 credits, rises to 2K for 4.5, and reaches 4K for 9.5. Start at 1K while working out the composition, then select a larger output when the destination needs it. All fourteen explicit aspect ratios are available, including wide banners and tall strips; choose the destination's shape before generating.

For lettering, spell out the exact words and say where they belong. For edits, describe the change and what should remain. Kuro returns PNG images and uses the existing prompt, reference, ratio and resolution controls; search grounding is not enabled. Nano Banana 2 remains under Older versions for work that already names it.

Describe what you want in your own words and Kuro writes the prompt for you. If you would rather write it yourself, this is what Nano Banana responds to.

Nano Banana follows a prompt faithfully, so state the scene plainly. It reads instructions the way an editor does: "place the product on the left third, keep the background out of focus" lands as written.

It renders short text legibly. Write any words that must appear exactly as they should be spelled, and say where in the frame they sit.

It is the family that best preserves a reference's identity — a face, a product, a frame from earlier in the sequence — so it is the one to reach for when a scene must match what came before. Say what each reference is for, and refer to them by their order ("Image 1 is the bottle, Image 2 is the shelf"). A saved Subject's picture reaches the model first, so your own references count on from there.

Its other reason to exist is the canvas. Choose the shape the destination actually is — a banner, a strip, a tall panel — instead of cropping a standard ratio afterwards. It frames 1:4, 4:1, 1:8 and 8:1 natively, which no other model in Kuro does.

A still that will open a clip is the opening shot, not a portrait: everyone who acts in frame, at the height and distance the action needs, with any text the clip must keep already in the picture.

The same advice covers Nano Banana 2 and Nano Banana 2 Lite. Their pages note what differs.

What it can't do

• Fourteen references on 2.1, seven on 2 and 2 Lite, a saved Subject's picture included. Past the last seat, references drop off the end of the list, so put the important one first.
• Resolution is 1K, 2K or 4K on 2.1 and 2, and each step up is a real step in cost. 4K should be a decision, not a default.
• Rewording to fix one detail re-rolls the whole picture. Edit the finished image with only the change.

Use something else when…

• You want painterly or stylised texture: FLUX.2 Pro.
• The frame holds dense small type, or a long instruction placing several objects: GPT Image 2.5 Flare.
• It is an everyday product or lifestyle shot at a standard ratio: Seedream 5 Lite.
• You want mass drafts with no references: FLUX Schnell.
• You need a vector file: Recraft V4.1 SVG.