Gemini uses Nano Banana (and its successor, Nano Banana 2) for all its native image generation and editing. While Google originally used models branded as Imagen, Gemini now relies entirely on the Nano Banana family. Click the plus icon next to the prompt box, pick Create image from More tools, then either click a template card in the Powered by Nano Banana 2 gallery or type a description in Describe your image. Submit and Gemini returns a 2x2 grid of variants you can hover to download, click to edit in the built-in canvas, or iterate on with a follow-up prompt or a reference image. Free on every Gemini plan. Verified May 2026.

How to Generate Images in Gemini with Nano Banana (2026)

Gemini's image generator looks like Imagen and is often still called Imagen by users out of habit, but the model behind the Create image tool, the +Image option, and the edit-this-image flows in Gemini today is Nano Banana, with Nano Banana 2 as its current successor. The Imagen brand is now reserved for Google's API-level direct-Imagen offering through Vertex AI, not for what runs inside Gemini chat. This guide walks the full Gemini image generation flow with screenshots from May 2026: opening Create image, picking a template or describing your own, saving and editing variants, and using image-to-image with a reference. Along the way, it shows where AI Toolbox (formerly Gemini Toolbox) complements Gemini with full-text search across every Gemini conversation (including every chat that produced a saved image) and multi-format export. AI Toolbox is part of the same unified Chrome install that also powers our ChatGPT and Claude modules.

What Is Nano Banana (and What Happened to Imagen)?

Nano Banana is Google's in-chat image generation model that replaced the Imagen-branded models inside Gemini. Nano Banana 2 is its current version and is labelled directly in the Create images template gallery ("Create with Nano Banana 2."). It powers every native image surface in Gemini chat: the Create image tool in More tools, the +Image option, hover-edit on returned images, and image-to-image generation from an uploaded reference. Imagen as a brand still exists, but only as Google's developer-facing model exposed via Vertex AI and the Imagen API. If you are clicking buttons in Gemini chat today, you are talking to Nano Banana, not Imagen.

This rename matters because the version you are using on a given day affects what the model can do (consistency across variants, follow-up edit fidelity, image-to-image quality). Gemini does not let you pick the model version manually inside chat: every Create image request inside Gemini today runs against the current Nano Banana family, with Nano Banana 2 as the default version surfaced in the templates UI.

Step 1: Open Create Image From the Plus Menu

Click the plus icon next to the Gemini prompt box, then pick Create image from the menu. The plus menu lists Upload files, Add from Drive, More uploads, and below the divider it shows the tools: Create image, Canvas, Deep Research, and More tools. Selecting Create image switches Gemini into image-generation mode and opens the Create images experience for this chat. You can also reach the same flow by typing "create an image of..." into a regular chat, but going through the menu gets you the template gallery as well.

This is the same menu you use for Canvas and Deep Research, so the muscle memory carries over. The Create image option is available on every Gemini plan, including free.

Step 2: Pick a Template or Describe Your Own (Powered by Nano Banana 2)

Inside the Create images modal, Gemini shows a 4x3 gallery of template tiles with the line "Create with Nano Banana 2." right under the heading. Templates include Monochrome, Colour block, Runway, Radiograph, Technicolour, Gothic Clay, Dynamite, Salon, Sketch, Cinematic, Steampunk, and Sunrise. Click any tile to seed your prompt with that style, then refine the rest in the Describe your image box at the bottom. The "Powered by Nano Banana 2" label is your confirmation that you are on the current model, not legacy Imagen.

If your style is not on the template list, skip the gallery entirely and write a free-form prompt in Describe your image. Templates are convenience, not a constraint. The same Nano Banana 2 model runs whether you start from a tile or from a blank prompt.

Step 3: Write a Concrete Prompt and Submit

Write a concrete prompt in Describe your image: name the subject, style, color palette, and any format constraints. A good prompt for the screenshots in this guide was: "Generate four images of a minimalist Chrome browser extension icon in purple gradient (#5e6292 to #c56fff). The extension is called AI Toolbox. Style: flat, modern, no text, no logo, friendly geometric shapes." After you hit submit, Gemini shows a small "Creating your image" loading indicator and a few seconds later returns a 2x2 grid of four Nano Banana variants in the chat.

Prompt habits that work well: pick one subject, one style, one palette, and one negative constraint (no text, no logo, square format). Vague prompts ("a nice logo") give vague results; specific prompts give variants you can actually choose between. If a variant is close but not right, keep iterating in the same chat (see Step 5) rather than starting over.

Search across every Gemini chat that ever made an image. Gemini's native sidebar search is limited to titles and recent chats, so finding the chat where you generated that one icon last month is painful. AI Toolbox (formerly Gemini Toolbox) adds full-text search across every synced Gemini conversation (including every chat that produced a saved image) with date, role, and exact-match filters, plus a Cmd+Shift+F (Ctrl+Shift+F on Windows) keyboard shortcut. Export any conversation as TXT, Markdown, JSON, or PDF for local archival. Free plan: 5 results per query. Premium: unlimited at $9.99 per month or $99 lifetime. Working across ChatGPT, Gemini, and Claude? All Access Lifetime ($149) covers all three modules and saves $148 versus three single-platform Lifetimes ($99 × 3 = $297). See the Gemini module →

Step 4: Edit or Save a Variant

Hover any variant in the 2x2 grid to reveal share and download icons; click the variant to open the built-in editor. The editor view shows the chosen variant on a transparent checker background with a small toolbar: a color row (black, red, orange, yellow, green, cyan, blue, magenta, gray, white) and two tools below it, Sketch (freehand drawing) and Text (overlay text labels). The top-right of the editor has Share, Download, and Save buttons. Use Sketch or Text to mark up the image, pick the brush color, and Save when you are done.

For a clean single-variant export, skip the editor entirely and use the hover download icon on the original grid, which downloads the raw Nano Banana output without your edits. Use the editor only when you actually want to annotate or composite something on top.

Step 5: Iterate With an Image-to-Image Reference

Attach an image to a follow-up prompt and Nano Banana 2 will generate new variants tuned to that reference. Use the plus icon next to the prompt box to upload a logo, photo, color swatch, or even a rough sketch, then write a prompt like: "Using this color palette and overall feel, generate four new minimalist icon variants for a productivity Chrome extension. Same square format, no text." Gemini returns a fresh 2x2 grid where every variant inherits the palette, line weight, or style of your reference. This is the closest Gemini gets to a one-shot brand-consistent image system inside chat.

For multi-step iteration in the same chat, you can also just reply with text: "Now make the icon look slightly more playful and less corporate, keeping the same purple gradient." Nano Banana 2 keeps context across the chat, so each follow-up regenerates the grid with that change applied without losing your original constraints.

How Nano Banana in Gemini Compares to Other AI Image Tools

Nano Banana in Gemini is closest in workflow to ChatGPT's image generation, but it differs in template surfacing, follow-up editing, and image-to-image flow. ChatGPT generates one image per prompt by default and lets you re-roll for variants; Gemini gives you a 2x2 grid every time. ChatGPT does not surface a labelled "powered by [model name]" template gallery; Gemini does. Claude does not generate images natively at all, and routes you to external tools.

Common Pitfalls With Gemini Image Generation

The most common failure modes are vague prompts, mismatched aspect ratios, and assuming Imagen is still the model. Vague prompts ("a nice purple logo") give generic Nano Banana output that looks like every other generic prompt run. Asking for an aspect ratio Gemini does not natively support (very wide banner crops, very tall mobile shots) often gets you square output regardless. And users still tell each other to "use Imagen in Gemini" when the actual model has been Nano Banana for months; the model selector is internal, so just trusting the templates UI label ("Create with Nano Banana 2.") is the safest reference point.

Other common pitfalls: re-rolling instead of refining (every fresh "regenerate" loses context, while a chat follow-up keeps it), using the canvas editor to do work you should ask the model to do (let Nano Banana 2 regenerate the icon in red rather than painting over it), and forgetting that uploaded reference images get fed back into the prompt, so an unrelated screenshot you attach will skew the variants. If results drift, start a new chat to clear the reference state.

Frequently Asked Questions

What model does Gemini use to generate images?

Gemini uses Nano Banana and its successor Nano Banana 2 for all native image generation and editing inside the chat. Google originally branded the in-chat image model as Imagen, but Gemini now relies entirely on the Nano Banana family. The Imagen brand is reserved for Google's API-level direct-Imagen offering, not for what runs inside Gemini chat.

What is Nano Banana 2?

Nano Banana 2 is the current image generation model that powers the Create image tool, the +Image option, and edit-this-image flows in Gemini. It generates four variants per prompt, supports follow-up edits in the same chat, and accepts a reference image for image-to-image generation. Google labels it directly in the Create images template gallery.

How do I generate an image in Gemini?

Click the plus button next to the prompt box, pick Create image from the More tools menu, then either click a template card in the Create images gallery or type a description in Describe your image. Submit and Gemini returns a 2x2 grid of Nano Banana variants. Hover any image for download or share options.

Is Gemini image generation free?

Yes. Image generation with Nano Banana is available on free Gemini accounts. Daily quotas, top-tier model access, and high-resolution exports are gated by your Gemini plan (free, AI Pro, AI Ultra), but the Create image tool itself does not require a paid subscription to use.

Can I edit a generated image in Gemini?

Yes. Click any returned variant to open it in the editor, where Sketch and Text tools let you mark up the image directly, and a color row lets you pick a brush color. You can also iterate by replying in the chat with a follow-up instruction (for example, make it more playful) and Nano Banana will regenerate the grid with that change applied.

Can I upload a reference image for Gemini to copy?

Yes. Attach an image to the prompt using the plus icon, then ask Gemini to generate variants in that style or based on that color palette. This image-to-image flow lets Nano Banana 2 produce on-brand visuals from a logo, photo, or rough sketch you already have.

How is Nano Banana different from Imagen?

Imagen is Google's general image model exposed via Vertex AI and the Imagen API for developers. Nano Banana (and Nano Banana 2) is the in-chat model Gemini uses for the consumer Create image experience. They share Google research lineage, but if you are using Gemini chat today, the model behind the images you see is Nano Banana, not Imagen.

Can I share or publish images Gemini generated for me?

Yes. Hover any generated image and choose the share icon to open Shareable public link to image, which gives you a copyable URL and one-click buttons for LinkedIn, Facebook, X, and Reddit. The same hover row exposes a download icon for saving the variant locally at full size.

Pair Nano Banana With AI Toolbox

Generating images in Gemini is fast; finding the chat that produced a specific image six weeks later is not. AI Toolbox (formerly Gemini Toolbox) indexes every message of every Gemini chat to IndexedDB in the background, so you can full-text search the prompt you used, the variant you described, or the style you settled on, with date, role, and exact-match filters and a Cmd+Shift+F shortcut. Export any conversation as TXT, Markdown, JSON, or PDF when you need a local archive of a long iteration thread. Free plan: 5 results per query. Premium: unlimited results at $9.99 per month or $99 lifetime. Cross-platform users can grab All Access Lifetime ($149), which covers the ChatGPT, Gemini, and Claude modules in one purchase. Add it from the Chrome Web Store; works on every Chromium browser (Chrome, Edge, Brave, Opera, Arc), not Firefox or Safari.

Conclusion

Gemini's image generation is no longer Imagen; it is Nano Banana, with Nano Banana 2 as the current default surfaced in the Create images template gallery. The whole flow is five clicks: plus icon to Create image, pick a template or describe your image, submit, hover or click any of the four variants to download or edit, and reply with a follow-up or a reference image to iterate. It is free on every Gemini plan and the variants are usable straight out of the grid for most icon, illustration, and concept work. When the chats stack up, pair Gemini with AI Toolbox so you can search and export every image-generation thread you have ever had.

Last updated: May 31, 2026