Table of Contents
A plain product photo on white background gets scrolled past in under a second. Nano Banana Pro turns that same photo into a scene — a café table, a rainy street, someone actually using the thing — and scenes get watched longer than product shots do. Here’s how to actually use it, not just try it once.
I. The Prompt Structure That Works
Skip long, adjective-heavy prompts. Use this order every time:
- Subject — the exact product, unchanged (upload it as a reference image, don’t describe it from scratch)
- Setting — one concrete location, not a mood (“marble kitchen counter,” not “elegant home”)
- Light — one specific condition (“late afternoon window light,” not “beautiful lighting”)
- Action — what’s happening, if a person is in frame
Example: “[reference: ceramic mug], on a wooden café table, morning light through a window, steam rising, hands wrapped around it.”
That’s it. Four slots, no filler. Vague prompts (“premium,” “aesthetic,” “eye-catching”) produce generic results because the model has nothing concrete to lock onto.
II. Why This Actually Moves Engagement
Two mechanisms, not vibes:
1. Context creates a reason to remember. A product with no setting is just an object; a product in a moment is a story fragment. Stories get recalled, specs don’t.
2. Custom beats stock, structurally. Stock photos are generic by definition — that’s why they’re free or cheap. A generated image built from your actual product reference will always match your palette and product shape more precisely than a stock library search ever will.
III. Batch Workflow For A Week Of Content
Don’t generate one image at a time. Do this instead:
1. Pick 3–5 SKUs you’re posting this week.
2. Generate 3 variants per product using the same prompt template, changing only the setting (café / outdoor / home).
3. Pick the strongest one per product, discard the rest. Don’t over-tweak — a slightly-off generation isn’t worth ten more attempts.
4. Save the winning prompt template. Reuse it next week with new products swapped in.
This turns a week of shooting into a 30-minute batch session.
IV. Keeping Your Feed Visually Consistent
The fastest way to make a feed look unbranded is switching lighting, angle, and mood every post. Fix it with two habits:
- Reuse the same reference image set for every generation — same product photography, same angle library.
- Lock one lighting style (warm/golden or clean/cool — pick one) and use it as a fixed phrase in every prompt for a full month.
If you’re running a launch across multiple posts, sequence them: unboxing → in use → gifted/shared. That gives the feed a narrative instead of a stack of unrelated shots.
V. Cost Reality
Standard-resolution generations are cheap enough to A/B test freely — generate two or three variants per post and pick the winner instead of guessing. Reserve higher-resolution output for anything going on a printed flyer, large banner ad, or a placement where users will zoom in. Don’t render every draft at max resolution; it wastes credits on images you’ll discard anyway.
VI. When To Use Nano Banana Pro vs. A Faster Model
- Everyday feed posts, single product, simple scene → a faster/lighter model is enough.
- Launch visuals, multiple reference images, layered composition, anything customer-facing at full size → Nano Banana Pro — it holds detail and reference accuracy better under complex prompts, which matters once you’re combining product + person + specific setting in one frame.
Test both on the same product once. The difference is usually obvious within one comparison, and you won’t need to re-test after that.
VII. Make It Repeatable, Not One-Off
The habit that actually sticks: pick one day a week, batch-generate the next week’s visuals in one sitting using your saved prompt template, and stop. Refresh the template itself every 4–6 weeks so the feed doesn’t go stale, but don’t rebuild it from scratch every time — that’s what kills the habit.
Start with one product you’re posting this week. Run it through Nano Banana Pro, compare it to your last plain product shot, and decide from there.