Rewritten October 2026. The first version of this post described Nano Banana as a text assistant. That was wrong, and this version replaces it.
Short answer: Nano Banana is the nickname for Google's Gemini image models. It can generate images from a prompt, but where it really earns its place is editing: you hand it a photo, describe the change in plain English, and it changes that one thing while leaving the rest alone. On LazySusan there are three versions, Nano Banana 2.1, Nano Banana 2 and Nano Banana Pro, and a standard 1K image costs between 41 and 161 tokens depending on which one you pick.
I've used it most weeks since the first version came out. Here's what I actually use it for, and when I reach for something else.
What is Nano Banana?
"Nano Banana" started as the codename Google used when it tested its first Gemini image model in public, and the name stuck. The current lineup:
| Version | Model ID | Best for | LazySusan tokens (1K / 2K / 4K) |
|---|---|---|---|
| Nano Banana 2.1 | gemini-nano-banana-2.1 | Everyday edits, the default | 41 / 61 / 136 |
| Nano Banana 2 | gemini-3.1-flash-image | Previous generation, still solid | 81 / 122 / 182 |
| Nano Banana Pro | gemini-3-pro-image | Text in images, fine detail, final assets | 161 / 161 / 288 |
LazySusan tokens work out to 600 per dollar of model usage, so a 1K image on Nano Banana 2.1 is roughly seven cents of your plan. The Starter plan's 1,000 monthly tokens covers about 24 of those.
You can pick any of ten aspect ratios, from 1:1 to 21:9, and 1K, 2K or 4K output.
1. Swap the background on product photos
This is the one I use most. Take a photo of a mug on your kitchen counter, upload it, and type: "Put this mug on a light oak table next to a window, soft morning light, keep the mug exactly as it is."
The mug stays the same mug, with the same logo and the same chip on the handle. Only the scene changes. For a small shop owner who can't afford a studio, that's the difference between phone snapshots and a listing that looks professional.
Tip: say what must not change. "Keep the label text identical" does more than you'd expect.
2. Remove things you didn't want in the shot
Power lines over a house photo, a stranger in the background of a portrait, a coffee cup on a desk you're about to use in a pitch deck. Upload the image and ask: "Remove the person on the right and fill in the wall behind them."
It handles this cleanly most of the time. Where it struggles is large removals in complicated scenes, like taking a car out of a busy street. If the first result smears, ask again with a narrower instruction rather than starting over.
3. Restyle a room before you commit to it
Thinking about painting the living room dark green? Photograph the room and ask for "the walls painted a deep forest green, same furniture, same lighting." Then try "warm white walls and a walnut floor" on the same photo.
I did this before repainting my office and it saved me two sample pots. It won't match a paint chip exactly, so treat it as a mood check, not a colour guarantee.
4. Fix a headshot without a reshoot
Uneven lighting, a busy background, a wrinkled collar. Ask for "the same person and expression, plain light grey background, softer even lighting." Keep the request modest. Nano Banana is good at preserving a face across small edits; ask it to change age, expression and pose all at once and you'll get someone who looks like a cousin.
5. Put readable text on an image
Older image models turned words into alphabet soup. Nano Banana Pro is one of the better models I've used for text inside images: a poster headline, a menu board, a label on a product mockup.
Keep the text short and put it in quotes in your prompt: a chalkboard sign that says "Fresh bread daily". For anything longer than a line or two, I still add the text in a design tool afterwards. This is the main reason to pay the extra tokens for Pro.
6. Build a consistent set from one image
If you need four images of the same product, same character or same mascot, generate or upload one you like, then keep editing from it: "same backpack, now on a hiking trail", "same backpack, on a train seat". Because each edit starts from the previous image, the details carry through far better than four separate prompts would.
7. Turn rough sketches into clean drafts
Photograph a whiteboard sketch or a napkin drawing and ask for "a clean flat illustration of this layout, same arrangement." It's useful for turning a rough wireframe or a floor plan into something you can show people. It won't give you a precise technical drawing, so don't use it for anything you'd measure from.
8. Make social images in the right shape
Because you choose the aspect ratio up front, you can make a 9:16 story, a 4:5 feed post and a 16:9 thumbnail from the same idea without awkward crops. Generate the 1:1 version first, then ask for the same scene in each format.
9. Quick concept images for decks and docs
When I need a picture for a slide that says "our customers are drowning in subscriptions", I don't want stock photography. A one-line prompt to Nano Banana 2.1 at 1K costs 41 tokens and is usually good enough on the first or second try.
For hero images where the look matters more than the edit, I switch models. More on that below.
Which Nano Banana should you use?
- Nano Banana 2.1 for almost everything. It's the newest, the cheapest per image, and the default in LazySusan.
- Nano Banana Pro when there's text in the image, when you need fine detail at 4K, or when it's a final asset you'll print or put on a homepage.
- Nano Banana 2 mostly if you liked how it rendered a particular style. It costs about twice what 2.1 does for similar work.
Nano Banana vs Midjourney and FLUX
All three are in LazySusan, and I use them for different jobs:
| Job | My pick | Why |
|---|---|---|
| Edit a photo you already have | Nano Banana 2.1 | Changes one thing and leaves the rest alone |
| Text on a poster or label | Nano Banana Pro | Most reliable lettering of the three |
| A striking image from scratch | Midjourney | Strongest sense of style and composition |
| Photorealistic product renders | FLUX 3 | Very clean, realistic lighting |
If you're starting from nothing and you care about the look, start with Midjourney or FLUX. If you're starting from a photo, start with Nano Banana. See the full model lineup for current versions.
Limits worth knowing
- One input image per edit. In LazySusan you edit one image at a time. Blending several reference photos into one result isn't supported there yet. (Developers can send up to three reference images per request through the LazySusan API.)
- Faces drift over many edits. After five or six rounds of changes, small differences add up. Go back to the original and make one combined request instead.
- Google adds an invisible watermark. Images from Google's image models carry SynthID, an invisible marker that identifies them as AI-generated. That's worth knowing if you're using the images commercially.
- It follows safety rules. Requests involving real public figures or certain edits of real people will be refused.
FAQ
Is Nano Banana free? Google offers limited free use in its own Gemini app. On LazySusan it's included in every plan and paid for with tokens: 41 tokens for a standard 1K image on Nano Banana 2.1, so about 24 images on the $8 Starter plan if you used your whole month on it.
What's the difference between Nano Banana 2.1 and Nano Banana Pro? 2.1 is newer, faster and about a quarter of the price for a 1K image. Pro is better at text inside images and fine detail, so it's the one to use for final assets.
Can Nano Banana edit my own photos? Yes, and that's its main strength. Upload a photo, describe the change, and it keeps the rest of the image as it was.
Is Nano Banana better than Midjourney? For editing existing photos, yes. For creating a striking image from scratch, Midjourney usually looks better. Many people use both, which is easy when they're in the same subscription.
Does Nano Banana work for logos? For early concepts, it's fine. For a final logo you'll want a vector file, and Recraft is the better tool because it can output SVG.
Can I use Nano Banana through an API? Yes. The LazySusan API offers Nano Banana 2.1 and Nano Banana Pro through one image endpoint, with up to three reference images per request, on prepaid credits that are separate from the app plans.
For more on how the image models compare, read 6 Things Nano Banana 2 Does Better Than Other Image AI, or see pricing to work out how many images your plan covers.