Nano Banana 2.1 Is Now in Karya: Sharper Images, 10 References, 4K Output

Nano Banana 2.1 is now in Karya. It improves on the Nano Banana 2 family in the places creators notice first: overall image quality, how closely it follows a detailed prompt, how well it keeps a character looking like the same person from one image to the next, and how cleanly it renders text inside an image.

What is new in Nano Banana 2.1

One model handles both jobs. Type a prompt with no images and it generates from scratch. Add reference images and the same model edits, restyles or combines them.

• Better quality and prompt adherence, so long, specific prompts land closer to what you described.
• Stronger character consistency, which matters when you build a set of images around the same person or mascot.
• More reliable text rendering for posters, labels, signs and social graphics.
• Up to 10 reference images in one edit.
• Three output sizes: 1K, 2K and 4K.
• A wide aspect ratio range, from square and portrait to 21:9 and tall or wide strips, plus an auto option.

How to use it in Karya

Open Image Studio and choose Nano Banana 2.1 from the model picker. For a fresh image, write your prompt, pick an aspect ratio and a resolution, and generate. To edit, add your reference images to the tray, then describe the change and say which reference is which. Karya shows the credit cost on the button before you commit.

A good habit for edits: name what must not change ("keep the product label exactly as shown") and change one thing per generation. That keeps faces, logos and layouts steady across a batch.

Pricing

Karya shows the exact credit cost before you generate, and unused credits never expire.

• Nano Banana 2.1 at 1K: 4 credits per image.
• Nano Banana 2.1 at 2K: 5 credits per image.
• Nano Banana 2.1 at 4K: 8 credits per image.

What to try first

Start at 1K to test a prompt or a reference set cheaply, then regenerate the winner at 2K or 4K for the final file. If you are producing a set of ad variations or character shots, keep the same reference images across the whole batch so the results stay consistent.