Nano Banana 2.1 vs Nano Banana 2: same prompts, side by side, and what to change in your code

Google released Nano Banana 2.1 (gemini-nano-banana-2.1) on October 6, 2026 as the update to Nano Banana 2 (gemini-3.1-flash-image). Before switching an image pipeline over, we ran both models on identical prompts and settings: one run each, no cherry-picking, default thinking for both.

Short version: 2.1 follows text prompts more literally and lays out infographics more cleanly. It was also slower in every test, because it thinks more by default. And it drops the 512px size.

The test

Setting	Value
Models	Nano Banana 2 and Nano Banana 2.1, same upstream
Thinking	Each model's default (Nano Banana 2: minimal; 2.1: medium)
Seed	None
Runs	One per prompt, first result kept

Left (or top) is Nano Banana 2, right (or bottom) is Nano Banana 2.1.

1. Text rendering: a chalkboard menu (4:3, 1K)

Both models spelled every word and price correctly. Nano Banana 2 added "1. 2. 3." numbering that the prompt didn't ask for and wrapped the dish names onto two lines. 2.1 kept the exact wording, and the layout is cleaner.

2. Infographic: the water cycle (1:1, 1K)

Both labeled all four stages correctly with no typos. Nano Banana 2 wrote longer explanations. 2.1 added a "THE WATER CYCLE" title, and its arrows read more clearly as a loop.

3. Wide banner: 8:1 at 2K

Google says 2.1 fixes tiling artifacts on wide and tall ratios at 2K and 4K. Neither image here shows visible seams, so one run can't confirm the fix either way.

4. Photo: a red fox (16:9, 1K)

Both are usable. The difference is mostly composition.

Speed

Test	Nano Banana 2	Nano Banana 2.1
Menu, 1K	15 s	25 s
Infographic, 1K	22 s	34 s
Banner, 8:1 2K	27 s	28 s
Photo, 1K	~10 s	~30 s

2.1 reported between 550 and 1,383 thinking tokens per image at its default medium level. If you need faster drafts, set the thinking level to minimal. Use high for dense layouts like infographics.

What Google's model card says

Google DeepMind scored 2.1 with blind human-preference tests, reported as Elo:

Test	2.1 (thinking)	Nano Banana 2	Nano Banana Pro
Text to image, overall	1050	990	935
Infographic design	1048	961	912
Multi-character consistency	1106	978	1011
Multi-reference editing	1066	988	989

These are Google's tests of Google's models, but 2.1 leads every row they published.

What to change when you switch

1. Model ID. On reAPI it is nano-banana-2.1.
2. Size. There is no 512px output any more. Use 1k, 2k or 4k.
3. Thinking. Decide whether the new medium default suits you, or set thinking_level to minimal or high.
4. Optional: Search grounding. 2.1 can ground an image in Google web search, image search, or both (google_search, google_image_search). Grounded results include search_suggestions, which Google requires you to display.

A complete submit-and-poll loop in Python:

import os
import time

import requests

API = "https://reapi.ai/api/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['REAPI_API_KEY']}"}


def generate(prompt, **params):
    task = requests.post(
        f"{API}/images/generations",
        headers=HEADERS,
        json={"model": "nano-banana-2.1", "prompt": prompt, **params},
        timeout=30,
    ).json()
    while True:
        time.sleep(4)
        task = requests.get(f"{API}/tasks/{task['id']}", headers=HEADERS, timeout=30).json()
        if task["status"] == "completed":
            return task["output"]["image_urls"]
        if task["status"] == "failed":
            raise RuntimeError(task["error"])


print(generate(
    'A cafe chalkboard menu with the heading "TODAY\'S SPECIALS"',
    aspect_ratio="4:3",
    resolution="1k",
    thinking_level="minimal",
))

The submit call returns a task id immediately. Polling doesn't cost anything, and an image usually takes well under a minute.

Price

Google's list price is per token. A 2.1 image costs $0.0336 at 1K and $0.0756 at 4K, plus thinking tokens. That is half of Nano Banana 2's image price ($0.067 at 1K), but 2.1's input and thinking tokens cost more. On reAPI, 2.1 is a flat $0.03 per image at 1K, 2K or 4K.

The prompts

1. Menu, 4:3, 1K: A cafe chalkboard menu with the heading "TODAY'S SPECIALS" and three items with prices: "Tomato Basil Soup $6", "Grilled Cheese $8", "Lemon Tart $5". Hand-lettered chalk, warm morning light, photorealistic.
2. Infographic, 1:1, 1K: A clean infographic explaining the water cycle with four labeled stages: Evaporation, Condensation, Precipitation, Collection. Arrows between stages, flat illustration, white background.
3. Banner, 8:1, 2K: A panoramic banner of a Mediterranean coastal town at sunset, colorful houses along the harbor, film photography. No text.
4. Photo, 16:9, 1K: A red fox sitting in a snowy pine forest at dawn, photorealistic. No text, no logos.

Written by the reAPI team. Model page and live price: https://reapi.ai/models/nano-banana-2-1 · API reference: https://reapi.ai/docs/nano-banana-2-1