Nano Banana 2 for E-Commerce Product Images: The Full Workflow

Product photography is expensive, slow, and annoyingly dependent on whether your photographer shows up. A decent studio shoot for a small product line can run you thousands of dollars before you’ve sold a single unit. Nano Banana 2 — the nickname the AI community slapped on Google’s Gemini 3.1 Flash Image generator — changes that math considerably. Launched February 26, 2026, it’s fast, it handles text on packaging without the usual garbled-letters disaster, and it keeps your product looking like itself across multiple shots. That last part is what makes it actually useful for e-commerce.

This tutorial walks through the full product-image workflow: from your first prompt to a polished, consistent set of shots ready for a product listing. You’ll get prompts you can copy and paste right now, plus the settings that separate results that look professional from results that look like you asked an AI to guess what your product looks like.

What You’ll Achieve

By the end of this tutorial you’ll have a repeatable workflow for generating product hero images, lifestyle shots, detail close-ups, and packaging renders — all with consistent product appearance, clean backgrounds, and text that actually reads correctly on labels. The same workflow applies whether you’re selling skincare, electronics, food, or furniture.

What You Need

Access to Nano Banana 2 (Gemini 3.1 Flash Image) comes through four routes: the Gemini app at gemini.google.com, Google AI Studio at aistudio.google.com, the Gemini API, or Vertex AI if you’re working in a Google Cloud environment. For most e-commerce sellers, the Gemini app or AI Studio is the fastest starting point. API and Vertex AI access matters if you’re building an automated pipeline — say, generating images for a product catalog at scale. Subject consistency across up to five characters is available across all access points.

Step 1 — Nail Your Base Product Prompt

The single biggest mistake people make with AI product photography is being vague. “A bottle of moisturizer on a white background” gives you a generic bottle. You need to describe your product the way a creative director would brief a photographer: material, color, shape, label details, and the mood you’re after. Here’s the base structure that works consistently well.

Professional product photography of a matte black glass serum bottle, minimalist white label reading "LUMÉ SERUM", placed on a smooth marble surface with soft diffused studio lighting, white background, 4K, sharp focus, no shadows on background

That prompt hits the key variables: material (matte black glass), label text (critical for Nano Banana 2’s improved text rendering), surface, lighting type, background, and resolution. The “no shadows on background” instruction prevents the muddy grey-shadow look that plagues AI product shots.

Step 2 — Generate Your Hero Shot Variants

A hero image is your primary listing photo — the one that shows the product clearly against a clean background. Generate at least three angles before picking one. Swap the angle descriptor while keeping everything else identical to maintain consistency.

Professional product photography of a matte black glass serum bottle, minimalist white label reading "LUMÉ SERUM", front-facing straight-on angle, white seamless background, soft even studio lighting, 4K resolution, commercial photography style, no props

Professional product photography of a matte black glass serum bottle, minimalist white label reading "LUMÉ SERUM", three-quarter angle view showing label and side profile, white seamless background, soft even studio lighting, 4K resolution, commercial photography style, no props

Professional product photography of a matte black glass serum bottle, minimalist white label reading "LUMÉ SERUM", top-down flat lay, white background, minimal soft shadow beneath bottle, 4K resolution, editorial clean aesthetic

Run all three prompts and compare label rendering across the outputs. Nano Banana 2’s text rendering is noticeably more reliable than earlier Gemini image models, but it’s not perfect — if the label text looks slightly off, add “text on label clearly legible, high contrast label design” to your prompt.

Pro tip ✅

Keep a “master product description” text block — the part describing your product’s physical appearance — and paste it identically into every prompt you run. Changing a single adjective can cause the generator to subtly alter the product shape or color. Consistency starts with identical input language.

Step 3 — Lifestyle Shots That Look Less Like Stock Photos

The dreaded “AI lifestyle image” problem is a model-with-uncanny-smile holding your product in a kitchen that doesn’t exist. Solve it by anchoring the scene with specific, grounded details and keeping humans peripheral or out of frame entirely.

Lifestyle product photography, matte black glass serum bottle labeled "LUMÉ SERUM" on a light oak bathroom shelf, beside a small white ceramic bowl of dried botanicals, morning sunlight coming from left window, warm neutral tones, editorial beauty photography style, 4K, shallow depth of field, product in sharp focus

Lifestyle product photography, matte black glass serum bottle labeled "LUMÉ SERUM" placed on a folded cream linen towel, flat lay composition, eucalyptus sprig beside product, soft natural light, spa aesthetic, overhead shot, 4K resolution

The botanical and linen props are doing heavy lifting here — they place the product in a lifestyle context without requiring the generator to render a human face, which is still where AI image generators tend to stumble. If you do need a person in the shot, keep them cropped to hands only: “woman’s hands with neutral nail polish gently holding the bottle” works far better than full-body or face requests.

Pro tip ✅

Real-time web grounding in Nano Banana 2 means the model has awareness of current visual trends. When describing your aesthetic, name a specific style rather than a vague mood — “Scandinavian minimalist beauty editorial” or “Korean skincare brand campaign” will pull from a real visual vocabulary the model recognizes, rather than averaging its way to blandness.

Step 4 — Product Detail and Close-Up Shots

Amazon, ASOS, and most major marketplaces now encourage or require detail shots showing texture, materials, and fine print. This is where Nano Banana 2’s 4K resolution and text rendering capability actually earns its keep.

Extreme close-up macro product photography of a matte black glass serum bottle cap, textured matte surface detail visible, soft studio lighting creating subtle surface sheen, no background distractions, 4K, commercial product photography, ultra sharp focus on surface texture

Close-up product photography of a white label on a matte black serum bottle, text clearly legible reading "LUMÉ SERUM / 30ml / Apply morning and evening", clean serif typography, white background, flat lighting, 4K, sharp text focus

The second prompt is a text-rendering stress test. If Nano Banana 2 renders “30ml” as “30nl” or scrambles the secondary copy, that’s normal — try adding “precise typographic label, all text correctly spelled and legible” and regenerate. For critical label text, generating five to ten variants and selecting the cleanest one is faster than iterating prompts endlessly.

Warning ⚠️

All images generated by Nano Banana 2 carry SynthID watermarks — Google’s invisible digital watermark embedded in the pixel data. These watermarks don’t affect how the image looks to human eyes, but they are detectable by verification tools. For most e-commerce use cases this is a non-issue, but if a client or marketplace ever asks about image provenance, you have your answer ready.

Step 5 — Packaging and Multi-Product Shots

If you’re selling a product line, you need shots that show the full range together. Subject consistency across multiple products is one of Nano Banana 2’s more practical features — describe each product clearly and it’ll keep the visual family cohesive across a single image.

Professional product photography of three matching matte black glass bottles, labeled "LUMÉ CLEANSER", "LUMÉ SERUM", and "LUMÉ MOISTURISER" respectively, arranged in a slight arc, white seamless background, even soft studio lighting, 4K, consistent product design language across all three bottles, commercial beauty photography

Flat lay product photography of a complete skincare set: matte black glass bottles labeled "LUMÉ CLEANSER", "LUMÉ SERUM", "LUMÉ MOISTURISER", arranged symmetrically on white marble surface, top-down overhead shot, minimal styling, 4K resolution, clean beauty editorial aesthetic, labels clearly legible

Multi-product prompts require more work from the model, and you’ll see more variation in quality between outputs. Generate more variants than usual — eight to ten — and expect to pick the best two or three rather than the best one.

Pro tip ✅

For a product line with consistent packaging design, build a “design system” in your prompt: always specify the same material, finish, label color, and typography style description. “Matte black glass, minimalist white label, clean sans-serif typography” acts as a visual system prompt that keeps all products in the same visual family across separate generation sessions.

Step 6 — Social Media Crops and Platform-Specific Variants

A product hero image for Amazon is not the same asset as your Instagram square or your Pinterest tall pin. Prompt for the platform from the start rather than cropping after — you’ll get better compositions and fewer decapitated product shots.

Square format product photography 1:1 ratio, matte black serum bottle labeled "LUMÉ SERUM" centered in frame, white background, soft studio lighting, Instagram feed aesthetic, 4K, product takes up 60% of frame with breathing room around edges

Vertical 4:5 format product photography, matte black serum bottle labeled "LUMÉ SERUM" with lifestyle props, cream linen background, warm natural light, Instagram feed post aesthetic, product positioned in lower two-thirds, negative space at top for text overlay, 4K

Note 💡

If you’re accessing Nano Banana 2 via the Gemini API or AI Studio, you can set aspect ratio as a parameter directly rather than describing it in the prompt. This gives cleaner results than trying to describe dimensions in natural language, especially for non-standard ratios like 9:16 for Stories.

Step 7 — Editing and Iteration Workflow

Nano Banana 2 supports conversational editing — you can follow up a generation with refinement instructions rather than rewriting the whole prompt from scratch. In the Gemini app, this works in the chat interface. In AI Studio and the API, it works through multi-turn conversation context. Use it for small adjustments: “make the background slightly warmer”, “remove the prop on the right”, “increase the depth of field blur on the background”. For bigger changes — different angle, different surface, different lighting style — rewrite the full prompt. Conversational edits work best for tweaks, not overhauls.

Avoid 🚫

Don’t stack too many conversational edits in a single session. After three or four rounds of “change this, adjust that”, the model tends to drift — subtle shifts in product color, label positioning, or lighting consistency creep in. If you’re more than three edits deep and still not happy, start fresh with a revised base prompt incorporating all your requirements from the start.

Building a Repeatable Production Pipeline

For sellers with more than a handful of SKUs, the Gemini API is where the workflow becomes genuinely scalable. You can template your prompts with product variables, feed them programmatically, and generate complete image sets for a product catalog without touching the Gemini app interface. Vertex AI adds enterprise-level controls — rate limits, usage tracking, and integration with existing Google Cloud infrastructure — which matters if you’re running a larger operation or building this into an internal tool for a team.

The practical result: a product launch that would have taken a week of back-and-forth with a photographer and retoucher can now take an afternoon. You won’t eliminate human photography entirely — some products genuinely need the real thing, and experienced photographers know angles and lighting tricks that no prompt can fully replicate yet. But for standard listing images, lifestyle variants, and social content, Nano Banana 2 is fast enough and good enough to make the math work in your favor.

promptyze

See Full Bio