How to Stop Google Gemini From Using Nano Banana (2026)

Short answer: there is no single switch that turns Nano Banana off. Nano Banana is Google's built-in image model inside Gemini, not an add-on you can uninstall. But you can stop it from kicking in — by controlling how you prompt, starting a clean chat, and (for developers) choosing the right model. Here's exactly how.

If Gemini keeps generating a picture when you wanted text, or edits a photo you only wanted it to describe, the problem isn't a rogue setting — it's how Gemini decides when to reach for its image model. Fix the trigger and the behavior stops.

Why Gemini keeps using Nano Banana

Nano Banana is Google's nickname for Gemini's image model (Gemini 2.5 Flash Image, with the newer Nano Banana Pro on top). Gemini routes a request to it based on intent: if your message looks like you want to create or change an image, it generates or edits one.

Two things push it toward image mode:

Image verbs in your prompt — "create," "generate," "make," "edit," "turn this into," "make it look like."
Conversation carryover — if earlier turns in the same chat involved an image (you uploaded one, asked for edits, or generated one), the conversation stays "near" image mode and the next reply often produces an image too.

So the goal isn't to disable anything — it's to stop sending the signals that trigger it.

Can you turn Nano Banana off completely?

No — not with a personal Gemini account. There's no toggle in settings that disables only the image model while keeping everything else. Plenty of pages promise a secret switch; there isn't one, and Google doesn't document a way to remove a core capability. What you have is control over when it fires, which in practice is what people actually need.

There's one exception, covered below: managed Workspace accounts, where an admin can restrict Gemini features centrally.

How to stop Gemini from generating or editing images

1. Tell it explicitly to stay in text

The most reliable fix needs no settings at all. Add a plain instruction:

"Answer in text only. Do not generate or edit any image."

When you upload a photo for analysis, be just as explicit:

"Describe what's in this image in text. Do not modify or recreate it."

Avoid the image verbs ("make," "create," "edit," "turn into") unless you genuinely want an image. Phrase requests as "describe," "analyze," "list," or "explain."

2. Start a fresh chat

If a conversation has already produced or edited images, that context lingers. Open a new chat and ask your text question cleanly — the routing resets and Gemini stops defaulting to image output.

3. Don't attach an image if you only want words

Uploading an image plus anything that sounds like a change ("can you fix the lighting here?") is the classic trigger for editing instead of describing. If you only want analysis, either say so explicitly (step 1) or describe the image in words rather than attaching it.

4. Developers: pick the model, not the magic

If you're hitting unwanted image generation through the Gemini API or Google AI Studio, you're in control of the model:

Call a text model (e.g., gemini-2.5-flash / a Pro text model) instead of the image model (gemini-2.5-flash-image, the one branded Nano Banana).
Don't request image output modality — if you don't ask for IMAGE in the response modalities, you won't get one.

Image generation in the API only happens when you select the image model or explicitly request image output. Choose a text model and the behavior disappears.

5. On a managed Workspace account, ask your admin

If your Gemini access comes through a work or school Google Workspace account, an administrator can enable or restrict Gemini features from the Admin console. If you need image generation disabled organization-wide, that's the lever — it's an admin control, not a per-user one.

"I actually just want to remove the Gemini watermark"

Some people search this because they don't want the Gemini logo or watermark on a Nano Banana image, not because they want to stop generation. Be aware of two layers:

A visible label/logo can sometimes be avoided by cropping, depending on the surface.
An invisible SynthID watermark is embedded in every Gemini-generated image by design. It can't be removed, and that's intentional — it's how AI-generated content stays identifiable.

So you can stop Gemini from making the image in the first place, but you can't strip the SynthID watermark from one it already made.

FAQ

Is there a setting to turn off Nano Banana in Gemini? No global setting on personal accounts. You control when it triggers through your prompts and by starting fresh chats. Only Workspace admins can restrict Gemini features centrally.

Why does Gemini edit my photo instead of describing it? Because the request read as an edit. Say "describe this image in text only, do not modify it," and avoid edit verbs like "fix," "change," or "make it."

How do I stop Gemini from generating an image mid-conversation? Start a new chat — image context carries over within a conversation, so a clean chat resets the routing.

Can I stop it through the API? Yes — use a text model rather than gemini-2.5-flash-image, and don't request image output modality.