Nano Banana 2: How to Make AI Images in Gemini

Nano Banana 2 is the nickname for Gemini 3.1 Flash Image, the image generator Google announced on February 26, 2026. It replaced the older Nano Banana Pro across all Gemini models—Fast, Thinking, and Pro—and Google pitched it as delivering Pro-level quality at Flash speed. If you use the Gemini app or website, this is the engine making your pictures.

Unlike tools that require design skills or complicated software, Nano Banana 2 works entirely through conversation. You type what you want to see, Gemini generates it, and you can ask for changes in the same chat. It’s free to start, works on your phone or computer, and every image carries a digital watermark so people know it was made by AI.

What Nano Banana 2 does

Nano Banana 2 turns text into images. Describe a scene—”a golden retriever wearing sunglasses on a surfboard at sunset”—and the model generates a picture in seconds. You can be vague or detailed; the AI fills in what you don’t specify.

It also edits. Upload a photo of your living room and ask Gemini to change the wall color, add a window, or swap the furniture. The model redraws the parts you mention and leaves the rest alone. You don’t need Photoshop or any design experience—just plain sentences.

Every image Gemini makes includes two invisible marks: SynthID, Google’s watermark that survives cropping and filters, and C2PA Content Credentials, metadata that records the file as AI-generated. Anyone with the right software can check those marks and see the image came from an AI, not a camera.

Which Gemini plan you need

Google’s Gemini plans page lists image generation with Nano Banana 2 on every tier, including the free one. Here’s what each plan gets you:

Google does not publish a fixed number of images per day for any plan. Limits vary and can change without notice. Free users will hit a cap sooner than paid subscribers, but the exact number depends on server load, your usage pattern, and Google’s policies at the time. If you run out, Gemini will tell you to wait or upgrade.

The “Redo images with Nano Banana Pro” feature on paid plans lets you refine an image Gemini already made—tweak colors, adjust composition, or regenerate parts—using the older, more precise Pro model. Free users can still ask for changes, but Gemini uses Nano Banana 2 for every step.

You can compare Google’s plans in detail at Is Gemini free? Google AI Plus, Pro and Ultra explained.

How to make an image with Nano Banana 2

Step 1: Open gemini.google.com in your browser, or launch the Gemini app on your phone. Sign in with your Google account if you haven’t already.

Step 2: Type a description of the picture you want. Be as specific or as loose as you like. Examples:

• “A cozy coffee shop with plants and string lights”
• “A red sports car parked in front of a modern glass building”
• “A cartoon cat reading a book under a tree”

Step 3: Press enter. Gemini will generate the image in a few seconds and show it in the chat.

Step 4: If you want changes, just keep talking. “Make the sky darker,” “Add a person in the background,” “Change the car to blue.” Gemini treats each message as an edit to the last image, so you don’t start over.

Step 5: When you’re happy with the result, tap or click the image to open it full-size, then save it to your device.

How to edit your own photos

Nano Banana 2 can also modify pictures you already have—a room you want to redecorate, a product photo that needs a different background, a snapshot where you’d like to change one detail.

Step 1: In the Gemini chat, click the attachment icon (a paperclip or plus sign, depending on your device).

Step 2: Upload the photo from your computer or phone.

Step 3: Describe what you want to change. “Remove the background and replace it with a beach,” “Change the wall color to light blue,” “Add a vase of flowers on the table.”

Step 4: Gemini will generate a new version with your edits. You can keep refining in the same chat—”Make the vase smaller,” “Move it to the left,” and so on.

Privacy note: Photos you upload go to Google’s servers for processing. Don’t upload pictures of other people without their permission, and avoid anything sensitive—medical records, IDs, private documents. Google’s privacy policy explains how they handle your data, but the safest rule is: if you wouldn’t email it, don’t upload it.

Tips for better results

Be specific about style. “Photorealistic,” “watercolor painting,” “3D render,” “pencil sketch”—Nano Banana 2 adjusts the look based on these cues. Without them, it picks a default style that may not match what you pictured.

Name the mood or lighting. “Warm sunset light,” “dark and moody,” “bright and cheerful.” Lighting shapes the whole feel of an image, and the model responds to these words.

Start simple, then refine. A short prompt—”a mountain cabin in winter”—gets you a starting point fast. Once you see it, you can add details: “Put smoke coming from the chimney,” “Add a snowman in the yard.” Iteration beats trying to write the perfect prompt on the first try.

Ask for variations. If Gemini’s first attempt is close but not quite right, say “Show me three versions of that” or “Try a different angle.” The model will generate alternatives without you rewriting the whole description.

Use reference images when editing. If you upload a photo and ask for changes, Gemini keeps the original composition and lighting intact, only altering what you mention. That consistency is harder to get from a text-only prompt.

What Nano Banana 2 won’t do

Google blocks prompts that ask for real people’s faces, violence, nudity, or anything that violates its generative AI use policy. If you hit that filter, Gemini will refuse and explain why. You can rephrase and try again, but the core restriction won’t change.

The model also struggles with fine text. If you ask for a sign or book cover with legible words, expect garbled letters. For anything that needs readable type—posters, logos, UI mockups—you’ll have better luck adding the text yourself in a separate editor after Gemini generates the image.

Hands and complex poses are still tricky. AI image models have improved here, but you may see extra fingers, twisted limbs, or odd proportions. If it matters, regenerate a few times or use the refine tools in a paid plan.

Comparing Nano Banana 2 to other tools

Nano Banana 2 sits in the middle of the pack: faster than the highest-end models, better than the cheapest free tools, and tightly integrated with Google’s ecosystem. If you already use Gmail, Docs, or Google Photos, Gemini is one click away.

ChatGPT with Images 2.5 is the closest rival. OpenAI’s generator is available on the free tier, handles prompts in a similar conversational style, and also lets you edit images by uploading them. Quality is comparable; speed is close. The choice often comes down to which assistant you prefer for text tasks. Read more at how to use the ChatGPT image generator.

Midjourney produces higher-fidelity art—more detail, more stylistic control—but it costs $10/month minimum, runs through Discord (not a chat app most people already use), and has a steeper learning curve. See our Midjourney review for the full breakdown.

Free-tier tools like Microsoft Designer and Canva’s AI features work without a subscription, but they limit you to a few images per day and push you toward templates and preset styles. Nano Banana 2 is more flexible and doesn’t force a layout on you.

For a broader comparison, check the best AI image generators and free AI image generators.

Common mistakes to avoid

Uploading other people’s photos without permission. Even if you only want to change the background or lighting, the original photographer or the person in the picture may not want their image fed into an AI. Ask first.

Passing off AI images as your own photography. Every Nano Banana 2 image carries the SynthID watermark and C2PA credentials. Claiming you took the photo yourself is easy to disprove and breaks trust. If you’re submitting work for school, a job, or a contest, check the rules—many now require disclosure or ban AI-generated content outright.

Assuming “free” means “unlimited.” Google can and does throttle free users when demand is high. If you plan to generate dozens of images in one sitting, expect to hit a wall. Paid plans raise that ceiling but don’t remove it entirely.

Ignoring privacy. Anything you upload or describe goes to Google’s servers. Don’t use Gemini to edit photos of documents, IDs, credit cards, or private medical images. If it’s sensitive, keep it off the cloud.

Frequently asked questions

Is Nano Banana 2 actually free?
Yes. Google includes it in the free Gemini plan, though you’ll hit usage limits sooner than paid subscribers. Google doesn’t publish exact daily caps, and they can change at any time. If you run out, Gemini will prompt you to wait or upgrade.

Can I use Nano Banana 2 images commercially?
Google’s generative AI terms allow commercial use of images you create, but you’re responsible for ensuring you don’t violate copyright, trademark, or publicity rights. The watermark and C2PA data stay in the file, so anyone can verify it’s AI-generated. Some clients or platforms ban AI art—check before you sell or publish.

How do I remove the AI watermark?
You don’t. SynthID is embedded in the pixels and survives cropping, resizing, and filters. C2PA metadata can be stripped with the right software, but doing so to deceive someone about the image’s origin may violate platform rules or laws in some places. The watermark is there for transparency; leave it.

Why does Gemini refuse my prompt?
Google blocks requests for real people’s faces, explicit content, violence, and anything that breaks its use policy. If you get a refusal, reword your prompt to remove the flagged element. The filter isn’t perfect—it sometimes blocks harmless requests—but appealing through the app doesn’t change the outcome. Try a different description or a different tool.

Can I train my own version of Nano Banana 2?
No. Google doesn’t release the model weights or code for Nano Banana 2. It’s only available through the Gemini app and API. If you want a model you can run and customize yourself, look at open-source options like Stable Diffusion or Flux—both require more technical setup but give you full control.

How does Nano Banana 2 compare to video generators?
Nano Banana 2 makes still images. If you want video, you’ll need a different tool—Google’s Veo 3.1, OpenAI’s retired Sora models, or alternatives like Seedance 2.5 and MiniMax H3. These are separate services with their own pricing and limits. Read the best AI video generators for a rundown, or see how to turn an image into a video with AI.

When to use Nano Banana 2

Nano Banana 2 shines when you need an image fast and you’re comfortable describing it in words. Blog illustrations, social media posts, brainstorming visuals for a project, mockups for a design you’ll refine later—all good fits. It’s built into a tool you may already use, it’s free to start, and it doesn’t require you to learn new software.

It’s not a replacement for a photographer when the stakes are high—headshots, product shots for a store, anything that needs to look unmistakably real and professional. The watermark, the occasional AI weirdness (strange hands, garbled text), and the fact that some clients simply won’t accept AI work all limit where you can use it. For those cases, hire a human or use a camera.

But for everyday image needs—turning an idea into something you can see, editing a photo without opening Photoshop, exploring concepts before you commit—Nano Banana 2 does the job. It won’t win awards, but it’ll save you time, and that’s what most people need from an AI tool.