Nano Banana for Product Photos: A Shopify Workflow That Works

Nano Banana is good enough for real listings now. The way most founders use it isn't.

Here's the short answer. Nano banana product photos hold up on a live Shopify page when you start from one clean, real photo of your product (the packshot), upload it as a locked reference, and change one thing per turn: background, then light, then props, then crop. The founders getting garbage are the ones typing a 90-word mega prompt and hoping the model keeps their label straight. It won't. Not reliably.

So this is the workflow I'd hand a founder today, plus the Amazon and Google Merchant Center rules that decide where these images are even allowed to go.

What is Nano Banana, and which version should you use?

"Nano Banana" is the nickname for Google's Gemini image models, and there are four of them now. According to Google's Gemini API image generation docs, the lineup is Nano Banana 2 (Gemini 3.1 Flash Image), Nano Banana Pro (Gemini 3 Pro Image), Nano Banana 2 Lite, and the original Nano Banana (Gemini 2.5 Flash Image), which Google now calls the legacy model.

For product work, the numbers that matter are in the same docs. Nano Banana 2 takes up to 14 reference images in one request, and up to 10 of those can be object images it keeps at high fidelity. It outputs from 512px up to 4K. Lite tops out at 1K, which is too small for a zoomable product gallery.

Google launched Nano Banana 2 on February 26, 2026, and its announcement leans on exactly the two things a store owner cares about: object fidelity across a workflow and legible text on things like packaging mockups. Every image also carries a SynthID watermark and C2PA Content Credentials. Remember that second part. It matters when we get to Google Shopping.

My take: use Nano Banana 2 as the default. Reach for Pro when the packaging is covered in small type.

For what it's worth, our own image pipeline runs on Nano Banana 2. I'm not neutral about the model. I'm opinionated about how it gets used.

Why does one mega prompt wreck your product?

Here's the thing. A mega prompt asks the model to invent a room, light it, style it, place your product in it and keep your product untouched, all in one pass. That's five jobs. The product is the job it quietly sacrifices.

What drifts first is always the same list. The label text picks up an invented letter. A mug handle gets a little thicker. A matte finish comes back glossy because the new lighting "wanted" a highlight. A practical Nano Banana editing guide from this month flags the same failures: shifted proportions, erased handmade texture, and maker's marks that look corrected but contain made-up characters.

A buyer doesn't forgive a pretty photo of the wrong product. They return it.

Google's own prompting guide points at the fix without saying it loudly. The Google Cloud prompting guide for Nano Banana says to "be explicit about what to keep exactly the same," and to refine images with follow-up prompts "in a conversational manner." Put those two lines together and you get the whole method: lock the product, then move one variable at a time.

How do you keep the product consistent in Nano Banana product photos?

The five steps, in order:

1. Shoot one real packshot on white, sharp and evenly lit.
2. Upload it and name what must never change.
3. Change the background only. Approve it.
4. Then light, then props, one turn each.
5. Check the product against the packshot before export.

Step 1: the packshot is the anchor. One real photo. Pure white background, product filling most of the frame, label facing camera, no heavy shadow. A phone on a tripod next to a window can do this. This single image does two jobs: it's your main image everywhere, and it's the reference every AI scene is built from. If the anchor is blurry, every scene inherits the blur.

Step 2: lock the product in words. Say it plainly in the first prompt: "Keep the bottle shape, the amber glass, the label text and the dropper cap exactly as in the reference image. Only change what I ask for." For a 2 oz serum with six words on the label, spell out those six words in quotes. Google's guide recommends quoting any text you want rendered, and that works for text you want preserved, too.

Step 3: background first, alone. "Place the bottle on a pale travertine bathroom shelf, morning light from the left." Nothing else. If the product survives this turn, you've got a stable base.

Step 4: one variable per turn. Softer light. Then a folded linen towel behind it. Then a tighter 4:5 crop for mobile. Each turn gets approved before the next one starts. If a turn breaks the label, go back one step and rephrase. Don't try to patch it forward.

Step 5: inspect the product before you judge the vibe. Zoom to 100%. Read the label letter by letter. Check the cap, the edge of the glass, the color against the packshot on the same screen. The scene can be gorgeous and still fail here.

Judge the product first. Judge the mood second. Most founders do it backwards.

Slower for ten minutes. Then faster, because you stop throwing most generations away.

Where can AI product photos go on Shopify, Amazon and Google Shopping?

This is the part that gets listings suppressed.

Destination	Can the main image be AI?	What the official rules say	My call
Shopify product page	No platform rule	None. The risk is trust and returns	Real packshot as image one, AI scenes in slots 2 to 5
Amazon main image	Risky	Must "accurately represent the product as a realistic, professional-quality image," pure white background (RGB 255, 255, 255), product at 85% of the frame	Real photo only
Amazon additional images	Yes	Must accurately represent the product and match the title. Photorealistic AI people need the contains-synthetic-performer tag	Good home for AI lifestyle scenes
Google Merchant Center	Yes	AI images allowed in image_link, additional_image_link and lifestyle_image_link with the IPTC DigitalSourceType tag kept intact	Real packshot in image_link, AI scenes as lifestyle images

The Amazon lines come straight from Amazon's product image guide in Seller Central, which also warns that a listing with no compliant main image may be pulled from search until you fix it. The Google lines come from Merchant Center's AI-generated content policy, which says outright: don't remove the embedded DigitalSourceType tag. And Google's image_link requirements add that the image has to match the product you're selling, with no promotional overlays covering it.

One practical trap. Plenty of image compressors and some Shopify apps strip metadata on upload. Before you send AI scenes to Merchant Center, check that the file you're submitting still carries its tag.

What does a bad AI photo actually cost you?

Run the math on a store like this, a hypothetical. A skincare brand selling that amber serum at a 1.5% conversion rate and a $78 average order value. That means revenue per visitor is $1.17. On 10,000 visitors, that's $11,700.

Now a drifted image slips into slot two. The label reads slightly wrong. A few careful buyers notice, a few more feel something is off without knowing why, and conversion rate slides to 1.2%. Same $78 average order value. Revenue per visitor is now $0.94. On the same 10,000 visitors, that's $9,360.

Clean product images $11,700

One drifted image $9,360

Conversion rate x $78 average order value x 10,000 visitors.

That's $2,340 a month lost to one image you could've caught at 100% zoom. The trust side of this is laid out in our breakdown of whether AI generated product images hurt Shopify conversion rate, and the short version holds: AI may change the room around your product. It may not change the product.

And here's the math on the other direction, because images are only one part of the page. On a bedding client's Cooling Bamboo Sheets, conversion rate went from 1.0% to 4.3% and average order value from $125 to $254. Revenue per visitor moved from $1.25 to $10.92. On 10,000 visitors, that's $109,200 instead of $12,500. See the full case study numbers. Real client numbers, not typical results, and not a promise of what your store will do.

Photos didn't do that alone. The page did: the words, the offer, the order of the proof. Good scenes make the page easier to believe. They can't make a mute page talk. Our guide to AI product pages covers the rest of the build.

Nano Banana makes the product look like it belongs in the buyer's life. The page still has to tell them why.

Where do lifestyle scenes earn their slot, and where do they hurt?

Lifestyle scenes pull their weight when the buyer needs to see scale or use: a throw blanket on a real sofa, a travel mug in a car cup holder. They hurt when they replace the honest shot. Our guide to whether lifestyle images increase Shopify conversion rate walks through which gallery slot each image type deserves.

Same logic as stock photography. A generic scene with no connection to your buyer is just decoration, and the argument in our stock photo audit applies to AI scenes word for word. Your buyer is a 34-year-old with a bathroom shelf and a skincare routine, so build that shelf. Not a marble spa that exists in no apartment.

Nano banana product photos: FAQ

Can you use Nano Banana for real product photos? Yes, for scenes built around a real photo of your product. Keep the main image a real photograph.

Does Nano Banana keep the product the same across different scenes? Much better than older models when you give it a reference and say what to keep, but check small label text against the packshot every time.

Can I use AI generated product images on Amazon? In the additional image slots, yes. The main image has to be a realistic, professional-quality image on pure white, and photorealistic AI people need the contains-synthetic-performer tag.

Does Google Merchant Center allow AI generated product images? Yes, as long as the IPTC DigitalSourceType tag stays in the file and the image matches the product you sell.

Which Nano Banana model is best for ecommerce product photos? Nano Banana 2 for most product scenes. Nano Banana Pro for packaging with lots of small text.

What to do next

Take your best seller. Shoot one clean packshot this week, then build exactly three scenes from it with the five steps above: one showing scale, one showing use, one showing the buyer's actual setting. Put them in slots 2, 3 and 4 and leave everything else alone for 14 days. If you're picking a tool to build the rest of the page around those images, read what to look for in an AI product page builder for Shopify first.

---

Book Your Profit Audit

Better photos only pay off when the page around them can close the sale. Book a profit audit and we'll show you exactly where your product page is leaking buyers, then how to rebuild a high-converting product sales page in less than 15 minutes.

Book Your Profit Audit →

Or go here to check it out → revenueflows.ai

P.S. One real packshot plus ten careful edits beats one hundred mega prompts. Lock the product, then let the model decorate the room.