Generic “product on white background” prompts produce generic results, and worse, results that don’t move clicks. A product ad needs four real decisions: composition (where the product sits in the frame), context (who uses it and where), lighting (key, fill, shadow direction), and a clear use-case angle. This tutorial walks through a four-angle template, picks the right 2026 model for the job, applies the brand-safety guardrails that keep your account alive, and runs the iteration loop that gets you a usable four-image listing in under an hour.
TL;DR
- Use a four-angle set: hero, in-use, lifestyle, detail. Lock one lighting style and one color palette across all four so it reads as a campaign, not four random renders.
- As of June 2026, Nano Banana Pro (Gemini 3 Pro Image) is the photorealism leader at roughly $0.13–0.14 per image, with native 4K up to 5632×3072 and 94–96% text-rendering accuracy. Seedream 5.0 Lite is the value pick at about $0.035/image. Midjourney V7 ($10–$60/mo) has gorgeous lighting but is weaker for literal commercial product shots.
- For your real SKU, use image-to-image with a reference photo (Nano Banana Pro accepts up to ~6–10 reference images). Text-only cannot reproduce your exact product.
- Brand-safety rules that get accounts banned if broken: no trademarks, no real-celebrity faces, no competitor products in frame, no AI-rendered text on packaging.
- Amazon main image must be pure white (RGB 255,255,255), 2000×2000px recommended, product filling 85%+ of the frame. The main image must accurately represent the real item.
Pick the right model first
The model decides 70% of your result. As of June 2026:
| Model | Approx. cost/image | Max resolution | Best for | Watch out |
|---|---|---|---|---|
| Nano Banana Pro (Gemini 3 Pro Image) | ~$0.13 API; in Google AI Pro $19.99/mo | 4K, up to 5632×3072 | Photoreal hero/detail, readable packaging text, reflective surfaces | Highest per-image cost |
| Nano Banana 2 (Gemini 3.1 Flash Image) | ~$0.067 | up to 4K | High-volume drafts, fast iteration | Slightly behind Pro on micro-detail |
| Seedream 5.0 Lite (ByteDance) | ~$0.035 | 2048×2048 | Value bulk runs, lifestyle scenes | Less precise text than Nano Banana Pro |
| Midjourney V7 | $10 / $30 / $60 per mo | ~2K | Mood, lifestyle, editorial lighting | Weaker for literal product reproduction |
Practical default: draft with Nano Banana 2 or Seedream 5.0 Lite to explore angles cheaply, then re-run the keepers at full quality on Nano Banana Pro for the final export. Nano Banana Pro’s text accuracy and glass/chrome handling earn the higher cost on hero and detail shots specifically. For the full field-by-field prompt anatomy — key light, fill, shadow direction, lens, brand mood — use our product image prompt guide as the underlying template for every angle below.
Who this is for
E-commerce sellers shooting their own listings, indie founders launching without a marketing budget, social-commerce sellers who need weekly fresh creative, and brand marketers filling an ad rotation without booking a studio every quarter.
When this is NOT the right tool
- Exact-product reproduction from text. AI cannot reliably copy your specific SKU. Use a real photoshoot, a 3D render, or background-removal on a real photo for the primary “this is what you receive” image. (Amazon’s 2026 rules explicitly require the main image to accurately represent the physical item.)
- Strictly regulated categories — supplements, medical devices, financial services — where misleading imagery can trigger a consumer-protection complaint.
Before you start
- Lock the product reference. Shoot 1–2 clean photos of your real product. Feed them as reference images (Nano Banana Pro and Seedream both accept multi-image reference); image-to-image gets you far closer than text alone.
- Decide output sizes up front. Amazon main image: 2000×2000px square, pure white background. Instagram feed: 1080×1350 (4:5). TikTok 9:16 stills: 1080×1920. Generate at 4K once, then crop platform variants.
- Decide brand visual direction. 2–3 hex colors, one lighting mood (clean studio / warm lifestyle / moody product-noir), one preferred prop category.
- Pre-list off-limits elements: real-celebrity faces, competitor logos, trademarked patterns, regulated claims.
Step by step
- Decide the ad angle first. The four-angle template: hero (product front and center), in-use (someone using it), lifestyle (product as part of a scene, not the focus), and detail (close-up of material, texture, craftsmanship).
- Hero prompt template:
[PRODUCT] centered, studio lighting with soft key from camera-left, gradient background in brand colors [HEX-1] to [HEX-2], slight contact shadow under product, clean catchlight on key surface, 4:5 ratio, 4K— attach your reference photo. - In-use prompt template:
[PRODUCT] being used by [PERSONA] in [SETTING], natural window light, lifestyle photography mood, candid mid-action moment, focus on product interaction not face, hands only. - Lifestyle prompt template:
[SETTING] with [PRODUCT] subtly visible on [SURFACE], soft morning light through window, shallow depth of field with background bokeh. - Detail prompt template:
macro close-up of [PRODUCT] material / texture / stitching, 100mm macro lens, soft directional light revealing texture, neutral background. - Lock the set. Keep the same lighting style and color palette across all four images; vary only subject and angle. This is the difference between “a set” and “four random images.”
- Stay brand-safe. Never prompt for trademarks, well-known character likenesses, real celebrities, or named competitor brands. These get accounts banned and can draw real legal action.
- Generate 6–8 per prompt, pick the best per angle, and save the winning prompts in one doc named by SKU.
A real run, start to finish
A new ceramic mug listing:
- Hero → 6 generations on Nano Banana 2 to scout composition → re-run the winner on Nano Banana Pro at 4K → pick 1.
- In-use →
hands holding mug, soft kitchen morning light, steam rising→ 6 generations → pick 1. - Lifestyle →
linen-textured table, partially-read newspaper, brand-color napkin, morning bokeh→ 6 → pick 1. - Detail →
macro on glaze texture, raking side light→ 6 → pick 1. - Final pass → open all four in one canvas, color-balance for unity, export at 2000×2000 (Amazon), 1080×1350 (Instagram), 1080×1920 (TikTok).
Time: about 90 minutes the first time including learning; 20–30 minutes per SKU once the templates are saved.
Quality check before you publish
- Set consistency: does it look like four photos from one shoot? Lighting and palette consistency is the test.
- Faces: any real-person faces? Swap to hands, silhouettes, or partial body. Full faces invite “is this real / paid?” complications and disclosure obligations.
- Packaging text: any AI-rendered label text? Even Nano Banana Pro’s 94–96% accuracy is not 100%. Replace label, dosage, and ingredient text with real text in Photoshop or Figma. AI text on labels reads as fake immediately.
- Hard surfaces: do reflective, glossy, and transparent surfaces look right? Glass, chrome, and jewelry are AI’s weak spots; expect 2–3× more rejects there. Nano Banana Pro handles them best.
- Brand-safe scan: no trademarks, no real-person likeness without rights, no competitor products visible in frame.
Brand-safety and platform disclosure (2026)
Get this wrong and you lose the account, not just the image.
- Amazon (2026): AI-enhanced backgrounds, lifestyle scenes, and infographics are allowed, but the main image must accurately represent the real product. Pure white RGB 255,255,255 background, product filling 85%+ of frame, no text/logos/watermarks/insets on the main image.
- TikTok (2026): any AI-generated content a reasonable viewer could mistake for real footage must be labeled; TikTok uses C2PA Content Credentials and invisible watermarking to auto-detect. Product-and-hands-only stills with no synthetic people generally avoid the “realistic person” trigger.
- Meta and Google: both increasingly require disclosure for AI-generated imagery of people. Safe practice: disclose, or keep your creative product-and-hands-only.
How to reuse this workflow
- Save the four prompts per category. All coffee mugs use the same four prompts with the product name as the only variable. Year two of your store becomes a 15-minute job per new SKU.
- Build a “rejected” library by reason: drifted logo, melted handle, fake packaging text. Naming the failure mode trains your prompt instinct.
- Re-test templates every 4–6 weeks. Model updates shift defaults; a “natural shadow” trick that you prompted explicitly may now be on by default.
- Keep one real reference photo next to every SKU’s prompt. Image-to-image will keep outperforming text-only as your reference library grows.
Common mistakes
- Reproducing your exact product from text alone. Use image-to-image with a reference photo, or limit AI to context shots.
- Every shot from the same angle. Listings need variety; the four-angle template exists for that reason.
- Adding real model faces. They invite uncanny-valley critique, model-rights questions, and platform disclosure rules. Use hands, silhouettes, or partial body.
- Generating text on packaging. Always add label text in an editor afterward.
- Inconsistent lighting across the set. Lock lighting and palette before generating.
- One-and-done. Generate 6+ per prompt; the first output is rarely the keeper.
FAQ
- Which AI model is best for product photos in June 2026?: For photorealistic hero and detail shots, Nano Banana Pro (Gemini 3 Pro Image) leads, with native 4K up to 5632×3072 and 94–96% text accuracy at roughly $0.13–0.14/image. Seedream 5.0 Lite (~$0.035/image) is the value pick for bulk lifestyle scenes, and Midjourney V7 is strong on mood but weaker for literal product reproduction.
- Can I sell using AI product images?: Yes, if you own rights to the product and follow platform and advertising rules. Avoid misleading representations of size, color, or features the product doesn’t have. Amazon’s 2026 rules require the main image to accurately represent the physical item.
- Will AI handle reflective surfaces?: Improving but inconsistent. For glossy metals, glass, and jewelry, expect 3–5× more retakes than matte products; Nano Banana Pro currently handles light refraction through glass best.
- Can I use AI to remove the background from a real photo instead?: Yes, and that’s often the right move for the primary listing image. Reserve generative AI for lifestyle and context variants where exact reproduction matters less.
- Do I have to disclose AI-generated ad images?: For imagery of people, increasingly yes — TikTok auto-detects via C2PA Content Credentials, and Meta and Google require disclosure. Product-and-hands-only stills generally avoid the realistic-person trigger.
- How long should the workflow take?: First SKU: about 90 minutes including learning. With saved templates: 20–30 minutes per SKU.