How to Keep Characters Consistent with Nano Banana 2.1

Google released Nano Banana 2.1 on 6 October 2026 as an update to the previous Flash image model. It improves visual quality, text rendering, and multi-turn character consistency while keeping Flash-level speed and cutting image costs roughly in half.

One standout feature is multi-image fusion. The model accepts up to 14 reference images in a single request. It maintains character consistency for up to four characters and object fidelity for up to ten objects. This makes it practical for storyboards, product campaigns, and brand assets that need the same person or item across scenes.

This guide shows how to use that capability in Google AI Studio and the Gemini API. All details come from the official model page and DeepMind model card.

What multi-image fusion supports

Nano Banana 2.1 (model ID gemini-nano-banana-2.1) takes text, images, video, and PDFs as input. For references you upload clear photos of the character or object you want to keep fixed.

Google states the model holds character identity across up to four people and object details across up to ten items. It also supports grounding with Google Web and Image Search when the prompt needs current facts. Thinking levels are configurable: minimal, medium (default), or high.

Default output is 1K resolution. You can request 2K or 4K. Wide and panoramic ratios no longer show tiling artifacts at higher resolutions.

Try it in Google AI Studio

Open Google AI Studio and select the Nano Banana 2.1 model. Start a new prompt.

1. Upload one or more reference images of your character. Use clear, front-facing or three-quarter views with consistent lighting when possible.
2. Write a short prompt that describes the new scene and explicitly asks to keep the same person. Example: “Show this person walking through a rainy Tokyo street at night, wearing a black coat. Keep the same face, hair, and body proportions.”
3. Set the aspect ratio and resolution if needed. For storyboards try 16:9 at 2K.
4. Generate. Review the result. Follow up in the same chat with a new prompt such as “Now place the same person inside a cafe, holding a coffee cup.”

The conversation history helps the model retain identity. Test with two or three turns before you commit to a full series.

If the face drifts, add a second reference photo that shows a different angle and repeat the instruction “match both reference images exactly for the character.”

Call the Gemini API

For production use, send the model ID gemini-nano-banana-2.1. Pass reference images as part of the input content.

A minimal Python example that generates a new scene while locking the character:

from google import genai
from google.genai import types
import base64

client = genai.Client()

# Load a reference image (replace with your file path or URL)
with open("character_ref.jpg", "rb") as f:
    ref_bytes = f.read()

response = client.models.generate_content(
    model="gemini-nano-banana-2.1",
    contents=[
        types.Content(
            role="user",
            parts=[
                types.Part.from_bytes(data=ref_bytes, mime_type="image/jpeg"),
                types.Part.from_text("Create a new image of this exact person standing on a mountain trail at sunrise. Keep face, hair, and build identical. Natural lighting, photorealistic.")
            ]
        )
    ],
    config=types.GenerateContentConfig(
        response_modalities=["IMAGE", "TEXT"]
    )
)

# Save the first image part
for part in response.candidates[0].content.parts:
    if part.inline_data:
        with open("output.png", "wb") as f:
            f.write(part.inline_data.data)

You can pass multiple images by adding more Part.from_bytes or Part.from_uri entries. The official image generation docs cover aspect ratio, resolution, and thinking level parameters.

Pricing starts at $0.0336 for a standard 1K image. Batch jobs cost half that rate. Confirm current rates on the Gemini API pricing page before you scale.

Tips for reliable consistency

• Use high-quality reference photos. Blurry or heavily filtered images reduce accuracy.
• Keep the character description short and repeated in every follow-up prompt. “Same person, same face” is clearer than a long backstory.
• Limit the number of characters in one image to four or fewer. Object references can go higher.
• For product shots, upload the exact item from multiple angles and ask the model to preserve logos, colors, and proportions.
• Generate at 2K when text must appear on clothing or signage. Higher resolution improves legibility.
• Enable thinking at medium or high if the scene is complex. It improves prompt adherence at a small latency cost.
• Test one edit with a single reference before you add more images. Confirm the model locks identity before you build a full campaign.

If wide banners previously showed seams on Nano Banana 2, retest those extreme ratios at 2K or 4K on 2.1. Google documented the tiling fix.

Watch a practical test

Several independent creators have stress-tested Nano Banana 2.1 against earlier models and other image systems. The video below compares character swaps and multi-reference results.

Related workflows

If you already build Android apps in Google AI Studio, the same model is available when you need on-brand visuals inside those projects. See our guide on building your first Android app with Gemini 4 Argon in AI Studio for the broader setup.

For small-business brand kits, Nano Banana 2.1 pairs with the new Gemini business landing page tools released on 20 October 2026. Upload your style once and reuse it across campaigns.

Conclusion

Nano Banana 2.1 makes character and object consistency practical at Flash speed and lower cost. Start in AI Studio with one clear reference photo, confirm the face holds across two or three turns, then move the same pattern to the API for production volume.

Check the official model page for the latest parameter names and pricing before you ship.

Sources

• Gemini Nano Banana 2.1 model page, Google AI for Developers: https://ai.google.dev/gemini-api/docs/models/gemini-nano-banana-2.1
• Nano Banana 2.1 model card, Google DeepMind: https://deepmind.google/models/model-cards/nano-banana-2-1/
• Image generation documentation, Google AI for Developers: https://ai.google.dev/gemini-api/docs/image-generation
• New Gemini tools for small business owners, Google Blog, 20 October 2026