Introduction

The Flow sighting was real. Google made Nano Banana 2.1 official on October 6: the Gemini API release notes list gemini-nano-banana-2.1 as generally available, a Google DeepMind model card went up the same day, and Google's launch post says it starts rolling out across the Gemini app, AI Mode in Search, Google AI Studio, Google Flow, Stitch, Google Ads, and Gemini Enterprise Platform.

It's a closed, API-only image generation and editing model, an update to Nano Banana 2 (Gemini 3.1 Flash Image). The headline for makers is simple: better edits and better consistency, at roughly half the per-image price of the model it replaces. And there's a clock attached, because Google has already deprecated Nano Banana 2 on the API.

What Changed

From Google's model page and release notes:

  • Better visual quality and realism at 1K, 2K, and 4K (1K is the default)
  • Fixed tiling artifacts on wide and panoramic ratios (1:4, 4:1, 1:8, 8:1) at 2K and 4K
  • Stronger text rendering and infographic layout
  • Up to 14 reference images, with character consistency for up to 4 characters and object fidelity for up to 10 objects
  • Grounding with Google Web Search and Image Search
  • Adjustable thinking levels: minimal, medium (the default), and high

Google's launch post leans on three wins in particular: visual design, mask-based editing, and subject consistency.

The Numbers

The model card compares Nano Banana 2.1 with thinking against Nano Banana 2 with thinking, in Elo-style ratings from Google's own human evaluations:

  • Multi-character consistency: 1106 vs 978
  • Mask and ink-based editing: 1049 vs 965
  • Multi-reference editing: 1066 vs 988
  • General editing: 1026 vs 938
  • Stylization: 1062 vs 991

These are vendor-run numbers, so treat them as a direction rather than a verdict until a public arena weighs in. Google is also upfront about the weak spots: small text can still blur at 1K, consistency isn't perfect, and left/right placement can get confused.

Pricing

On the Gemini API paid tier (standard, not batch):

  • 1K image: $0.0336 (Nano Banana 2 is $0.067)
  • 2K image: $0.0504 (Nano Banana 2 is $0.101)
  • 4K image: $0.0756 (Nano Banana 2 is $0.151)
  • Batch mode halves those again, to $0.0168 for a 1K image
  • There's no free API tier for this model, but you can try it in Google AI Studio

Every image carries a SynthID watermark.

The Deadline

This is the part to put in your calendar. The release notes say gemini-3.1-flash-image (Nano Banana 2) is deprecated and will shut down on October 29, 2026, and Google tells developers to migrate to gemini-nano-banana-2.1. If you have an app, bot, or workflow calling Nano Banana 2 by model ID, you have about three weeks to swap it.

Where to Run It Today

  • Gemini API, model ID gemini-nano-banana-2.1
  • Google AI Studio for quick tests
  • Google Flow, where it sits alongside Nano Banana Pro and Nano Banana 2 Lite
  • The Gemini app and AI Mode, as the rollout reaches your account
  • Vercel AI Gateway, as google/gemini-nano-banana-2.1

Try It

A minimal Python call using Google's Interactions API, straight from the image generation guide's pattern:

from google import genai
import base64

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-nano-banana-2.1",
    input="A cozy rooftop greenhouse cafe at dusk, warm string lights, rain on the glass",
    response_format={
        "type": "image",
        "mime_type": "image/jpeg",
        "aspect_ratio": "16:9",
        "image_size": "2K",
    },
)

with open("rooftop_cafe.jpg", "wb") as f:
    f.write(base64.b64decode(interaction.output_image.data))

To test the masked-edit upgrade, attach a photo and use Google's semantic masking template. Swap the bracketed parts for your own:

Using the provided image, change only the [specific element] to [new element/description]. Keep everything else in the image exactly the same, preserving the original style, lighting, and composition.

Original Source

Conclusion

Nano Banana 2.1 isn't a giant leap in name, but it's the kind of update makers actually feel: edits that stay inside the lines, characters that keep their faces across a set, and a bill that's about half what it was. Run your favorite masked-edit test on it this week, and if anything of yours still calls Nano Banana 2, migrate it before October 29.

Want to compare it against the rest of the family? Nano Banana Pro is already on Gen → https://artrealmai.com/gen?utm_source=magazine&utm_campaign=gen&utm_content=nano-banana-2-1-ga-gemini-api-half-price

—Aurelia ♡