Nano Banana for e-commerce works well when it starts from a real photo of the product: it's Google's family of image models (Nano Banana Pro, Nano Banana 2 and 2.1) that takes several reference images and edits a scene without starting over. It's less reliable where e-commerce is strictest: small text, labels, logos. This guide covers what each version does, where it delivers on product pages and how to handle the limits.
What Nano Banana is and which versions exist
"Nano Banana" is the name Google uses for its Gemini models that generate and edit images. In October 2026 three versions matter for e-commerce.
| Version | Model | What matters for e-commerce |
|---|---|---|
| Nano Banana Pro | Gemini 3 Pro Image | Up to 14 images blended, consistency for up to 5 people, 2K and 4K output, more accurate in-image text |
| Nano Banana 2 | Gemini 3.1 Flash Image (Feb 2026) | Features close to Pro at Flash speed, 512 px to 4K, translation and localisation of in-image text |
| Nano Banana 2.1 | gemini-nano-banana-2.1 (6 Oct 2026) | Updates version 2: up to 14 references (4 characters, 10 objects), 1K-4K, improved text |
Sources: Google, Nano Banana Pro, Google, Nano Banana 2, Gemini API, Nano Banana 2.1. On prompt adherence alone, Nano Banana Pro ranked first on the Lumenfall leaderboard updated on 28 July 2026 (Lumenfall source).
Where Nano Banana delivers on product images
Editing with references
Editing is the strength: upload the product photo, one of the background or location, maybe a lighting reference, and ask for a new scene. The model keeps the product and changes the rest. With up to 14 input images you can combine product, props and style in one pass (Google source).
Scene variants for the same listing
One shot becomes several settings: bathroom, kitchen, desk, outdoors. Useful for galleries and ads, where the same item needs several contexts. More on this in AI lifestyle product photography.
Consistency across a collection
Character and object consistency from one image to the next has improved with each version, and that's what makes a catalogue look like a single set. Google itself notes, though, that character consistency across edits "may vary" (Google source).
Localisation
Nano Banana 2 translates and localises text inside an image (Google source): handy for banners and campaign graphics in several languages. For text on the product packaging, see below.
The limits: labels, small text and logos
This is the part that decides whether an image actually goes live. In its notes on current limitations Google writes that rendering small text, fine details and accurate spellings "may not work perfectly", that multilingual text can contain grammar mistakes and that complex edits such as blending or lighting changes can produce unnatural artifacts (Google source).
For an online store, that means:
- Ingredients, dosages, claims and warnings on the label aren't generated: they're put back from the original artwork file.
- The logo is checked letter by letter, even when it looks right at a glance.
- Product colour is compared with the real shot: it tends to drift after a lighting change.
- Reflections and transparency (glass, metal, liquids) are where an image loses credibility.
The full method is in how to keep product, logo and packaging true in AI images.
"Nano Banana moves the product into a new scene in seconds. Checking it's still your product still takes a human eye."
A workflow that holds up
- Start from real photos. One clean shot per view, even taken in-house, beats any description.
- Write the prompt like a brief. Subject, background, light, framing, format. Tested structures are in our guide to product photo prompts.
- Generate less, choose well. Four checked variants beat forty to scroll through.
- Compare with the product. Shape, proportions, colour, logo, label: one check per image.
- Rebuild text in post. Labels and logos come back from the original vector file.
- Export per channel. Ratios, file sizes and backgrounds differ between your store, Amazon and Zalando.
When to do it yourself and when not to
For a few scene variants on products without critical text, using Nano Banana directly in the Gemini app or AI Studio can be enough. When there are dozens of listings, products carry regulated labels or you need visual direction consistent with the brand, the work shifts from generating to checking. That's where we come in: for our AI e-commerce images we use Nano Banana alongside other models, chosen per shot, and check every image against the real product. Prices start from €20 per image, with a 50-image package at €1,490 and delivery in 5-10 working days.
Frequently asked questions
Is Nano Banana good for e-commerce product photos?
Yes for editing and scene variants starting from real photos. It needs checking on small text, labels and logos, which Google itself flags as areas where the model can get things wrong.
What is the difference between Nano Banana Pro and Nano Banana 2?
Nano Banana Pro (Gemini 3 Pro Image) is the most capable model in the family. Nano Banana 2 (Gemini 3.1 Flash Image) brings similar features at Flash speed; version 2.1 from 6 October 2026 updates it.
How many reference images does Nano Banana accept?
Up to 14 images per request. For Nano Banana 2.1 the documentation states consistency for up to 4 characters and 10 objects.
Can Nano Banana render product labels correctly?
Text has improved a lot, but Google warns that small text and exact spelling may not be perfect. Ingredients, claims and warnings should be rebuilt from the original artwork file.
Can I use Nano Banana images on Amazon or Zalando?
The marketplace rules apply as for any image: technical specs and a product true to what the customer receives. Also check any transparency obligations for AI content.