Nano Banana 2.1 Explained: What Changed & How to Use It

Nano Banana 2.1 is Google's model for generating and editing images, generally available in the Gemini API since October 6, 2026. It's an update to Nano Banana 2, and Google pitches it as the more efficient counterpart to Nano Banana Pro rather than its replacement: the speed and price of a Flash model, with better image quality, prompt adherence, character consistency and text. Developers call it gemini-nano-banana-2.1, and Google's model card says it's built on Gemini 3.6 Flash.

In Spicy Studio, the Image Lab opens with Nano Banana 2.1 selected. You don't need a Google login or an API key: write a prompt, add reference images if you have them, and pay 12 credits for a 2K image or 18 for 4K from the same balance as every other model.

These changes come from Google's release notes and docs. The quality claims and scores are Google's own, from its internal tests published in October 2026.

Text and infographics. The biggest gain on paper. On Google's infographic-factuality test, 2.1 with thinking scores 0.521, Nano Banana 2 0.179 and Nano Banana Pro 0.265. With thinking turned off, 2.1 falls to 0.328.

A medium thinking level, used by default. Google's API lets you set how much the model reasons before it renders: minimal, medium or high. Nano Banana 2 offered only minimal and high and defaulted to minimal, so 2.1 reasons more out of the box, trading some latency for quality.

Characters that hold across turns. Google reports better multi-turn character consistency, so a face should survive a chain of edits more often. The limit stays at 14 reference images, with room for four consistent characters and ten objects.

Cleaner extreme formats. Panoramic and very tall ratios such as 8:1 and 1:8 arrived with Nano Banana 2; 2.1 fixes the tiling artifacts they produced at 2K and 4K.

No more 512px. Output starts at 1K, with 2K and 4K above it.

Cheaper output, pricier input. On Google's API a 2K image costs about half what it did on Nano Banana 2, while input tokens cost three times as much. Nano Banana 2.1 price has the full rate card.

Grounding in Google Web and Image Search isn't new; Nano Banana 2 already had it. On release day, Google also deprecated Nano Banana 2's API model, gemini-3.1-flash-image, without announcing a shutdown date.

Google's model card is candid about weak spots worth knowing before you plan a project around 2.1:

- Small text can come out blurry, especially at 1K, and long paragraphs still break down. - Character consistency between references and output is good, not perfect. - It occasionally confuses left and right. - In rare cases an edit keeps the subject's original pose. - It can hallucinate details, and Google mentions occasional slowness or timeouts.

So keep on-image text to one short line, generate at 2K or 4K (the only sizes the Image Lab offers for this model), name the hand or side you mean and check it, and reuse one face reference instead of editing the last output over and over.

Open the lab. The Image Lab loads with 2.1 already chosen. If you've switched away, it's first in the model list.

Write the shot, not a word cloud. Subject, setting, light and framing, in that order, usually beat a long run of style words. Put any words that must appear in the image inside quotes.

Attach references when you have them. Add as many as 14 images, such as a face, a product, a garment or a location, and point to them by number in the order you attached them: “the jacket from image ref 2”. They don't change the price. Without them, 2.1 works from text alone.

Pick size and shape. 2K costs 12 credits and 4K costs 18. There are ten aspect ratios from 1:1 to 21:9, including 9:16 for vertical video and 4:5 for feed posts, saved as PNG or JPEG.

Keep going from the result. Open a finished image and choose Edit Image to send it back as a reference for the next change, Upscale to take it to 4K or 8K in Image Upscale (10 or 12 credits), or Animate to open it in the Video Lab with Seedance 2.5 selected. To remove one object instead of re-prompting, the Eraser takes a brushed mask for 8 credits.

A poster with a real headline. Try: “Gig poster for a Friday jazz night in a narrow brick bar, a brass saxophone on a stool under one warm spotlight, the words ‘LATE SET’ in tall cream letters across the top, grainy risograph texture.” Use 2:3 at 2K and read every letter.

A product in a new setting. Photograph the real item on a plain surface, attach it as image ref 1 and ask: “The ceramic mug from image ref 1 on a wooden windowsill at sunrise, a thin curl of steam, glaze color and logo unchanged, shallow depth of field.” Use 4:5 or 1:1 for shop listings and check the logo against the real mug.

A first frame for a short clip. Generate a 9:16 still with hands and anything that will move fully in frame, then press Animate. A 5-second Seedance 2.5 clip costs 50 credits at 480p or 110 at 720p. The playbook's first-frame lesson shows the whole routine, including its Nano Banana 2.1 enhance pass.

For a character who has to appear again and again, the Nano Banana 2.1 prompts guide has recipes that keep one face across scenes and edits.

Some 2.1 controls exist only in Google's products. Through the Gemini API and AI Studio you can set the thinking level, ground a prompt in Google Search, render at 1K and use the panoramic ratios; the Image Lab doesn't show those settings today. In the Gemini app, 2.1 is reserved for Google AI Plus, Pro and Ultra subscribers, while free accounts get Nano Banana 2 (as of October 2026). Google also runs it in AI Mode in Search, Google Ads, Flow and Stitch.

What Spicy Studio adds is everything around the model: Nano Banana 2, Nano Banana Pro and GPT Image 2.5 on the same credits, plus the video models, with nothing to set up. GPT Image 2.5 is worth testing when a layout depends on lots of small text. When to switch between the Nano Banana models is covered in Nano Banana 2.1 vs Nano Banana 2 vs Pro, and the Nano Banana Pro guide looks at Pro on its own.