TL;DR
A good product image prompt is a photographer’s brief, not a wish. Name the material, key/fill/shadow lighting, background, lens, and aspect ratio, then lock those parameters and only change the seed. As of June 2026 the practical default is Gemini Nano Banana Pro (Gemini 3 Pro Image) for accurate labels and a consistent set, Midjourney v7 with --oref for stylised hero shots, and ChatGPT Images 2.0 when you want conversational edits inside a chat. Use AI for hero, lifestyle, and context shots; shoot the spec gallery on a real phone.
The task
You sell on Etsy, Amazon, Shopify, or your own store, and a studio shoot is out of budget. You need images that look like a small brand sweated the details: proper lighting, a clean background, the product at a flattering angle. AI image tools deliver this now, but only if your prompt specifies what you would tell a photographer — light direction, fill, shadow softness, background, lens choice. A one-line prompt (“nice photo of my candle”) gets you a generic stock look. A briefed prompt gets you a set.
What changed in 2026: AI can do labels now
The old rule was “AI can’t render text, so keep labels off the product.” That is no longer true. As of June 2026, Nano Banana Pro (Google’s Gemini 3 Pro Image model) renders legible, multilingual label text directly in the image, which removes the post-production text overlay step for many packaging shots. Two caveats still hold:
- It renders readable text, not your exact logo lockup at vector precision. For a regulated label or a registered mark, generate the scene, then composite your real label on top.
- It still cannot reproduce your specific product 1:1 — it makes a convincing look-alike. For the spec gallery where 1:1 accuracy is required, use a real photo.
So the division of labour is: AI for hero, lifestyle, and context; real camera for the accuracy-critical spec shots.
Pick the right tool for the shot
| Tool (June 2026) | Best for | Strengths | Cost to start |
|---|---|---|---|
| Nano Banana Pro (Gemini 3 Pro Image) | Packaging with real label text, consistent sets | Accurate text, reference-driven brand consistency, native 1K with 2K/4K upscale | Free tier in the Gemini app; API ~$0.134/image at 1K–2K |
| Midjourney v7 | Stylised hero shots, mood, cinematic lighting | Best material/atmosphere rendering; --oref keeps the same object across scenes | Basic $10/mo (~3.3 fast GPU hrs); Standard $30/mo |
| ChatGPT Images 2.0 | Conversational edits, quick iterations | Consistency across up to 8 images per prompt; edit by chatting | Included in ChatGPT Plus $20/mo (~200 images/day) |
| Photoroom / e-commerce tools | Background swaps on a real product photo | Trained on e-commerce; ghost-mannequin, staging templates | Free tier; paid plans from ~$10/mo |
A safe default workflow: shoot or generate one clean hero, then use Nano Banana Pro or ChatGPT to spin out the matching set so all images share one look.
What to put in the prompt
- Product name and a one-line description
- Material and finish (matte ceramic, brushed brass, polished oak)
- Brand mood (calm and quiet, bold and saturated, technical and clinical)
- Background (gradient that supports the product colour, marble, linen, on-figure, outdoors)
- Lighting direction (most images need a key light, a soft fill, and one subtle shadow)
- Use case (hero shot, on-figure lifestyle, scale reference, packaging close-up)
- Aspect ratio for the target platform (1:1 for Instagram, 4:5 for Etsy listings, 16:9 for banners)
Copy-ready prompt (works in any of the three tools)
Generate a product photograph.
Product: [name and one-line description]
Material and finish: [list]
Brand mood: [one line]
Background: [gradient that supports the product colour / marble / linen / outdoors]
Lighting:
- Key light: upper-left, 45 degrees, soft
- Fill light: low-power right, soft
- Shadow: subtle drop shadow on flat ground, not floating
Lens / framing: 50mm equivalent, front-three-quarter angle
Style reference: [2-3 photographers or brands]
Aspect ratio: [1:1 / 4:5]
Constraints:
- No invented brand text on the product (we composite the real label in post)
- No human hands unless the brief asks for lifestyle
- No floating product — it must sit on a believable surface
- Keep the label area clean and undistortedMidjourney v7 version of the same brief (parameters go at the end of the line):
matte ceramic mug, calm minimalist brand, on warm linen, soft key light upper-left
45deg, low fill right, subtle ground shadow, 50mm front-three-quarter, studio still
life --ar 4:5 --s 200 --v 7 --oref [URL of your hero image]For a lifestyle variant in any tool: “Same setup, but add a soft-focus human element (a hand pouring or holding) for context. The hand is out of focus and is not the subject.”
How to run the set so it stays consistent
Generate 4–8 images with the prompt parameters held constant and only the seed changing — same prompt, different seeds, not different prompts. In Nano Banana Pro, feed your chosen hero back in as a reference image so the model keeps the materials and lighting. In Midjourney, pass that hero with --oref so the object stays the same across scenes. Pick the strongest result, then iterate the rest by tightening exactly one parameter at a time.
QA checklist before you publish
- The product sits on a believable surface (no floating)
- Shadow direction matches the stated key light (light upper-left means shadow bottom-right)
- The background supports the product instead of competing with it
- Material reads correctly (matte should not look glossy)
- Every word of text in the image is intended — no invented brand names
- The set reads as one brand: 8 images that look like one shoot, not 8 styles
Common mistakes
- Busy background. AI defaults to dramatic if you do not specify. State the background plainly.
- Wrong shadow direction. Write “shadow falls bottom-right” when the key light is upper-left.
- Inconsistent set. Different lighting per image looks like different brands. Lock lighting and lens.
- Letting AI invent labels. Even Nano Banana Pro will write plausible-but-fake brand text unless told to leave the label area clean for your real composite.
- Skipping a real reference. Pull one photo you like and tell the AI to match its lighting and angle.
- Brute-forcing volume. Generating 50 and fishing for one wastes credits; fewer generations with intent wins.
FAQ
- Can AI exactly reproduce my product? No. It makes a convincing look-alike, not your exact unit. Use AI for hero, lifestyle, and context; shoot the spec gallery with a phone and a light.
- Can AI put real text on the label now? Yes, mostly. As of June 2026, Nano Banana Pro renders readable label text. For a registered mark or a regulated claim, generate the scene and composite your real label on top for vector-perfect accuracy.
- Which tool should I start with? Nano Banana Pro for packaging and consistent sets (free tier in the Gemini app), Midjourney v7 for stylised hero shots ($10/mo Basic), ChatGPT Images 2.0 if you prefer editing by chatting (in Plus at $20/mo).
- How do I keep 8 images on-brand? Generate the hero first, then feed it back as a reference image (Nano Banana Pro) or pass it with
--oref(Midjourney) so the materials, lighting, and object stay constant. - What about regulated categories (skincare, supplements, food)? Visuals can imply claims (medical effect, “organic”, before/after). Treat the image like ad copy and clear anything implied with whoever reviews your claims.
- What aspect ratio per platform? 1:1 for Instagram grid and Amazon mains, 4:5 for Etsy listings and Instagram portrait, 16:9 for site banners and ads.
Related
- AI product photo tutorial — end-to-end workflow
- Skincare product image prompts — category-specific prompts
- Jewelry product image prompts — reflective / small detail products
- Tech gadget product image prompts — gadgets and electronics
- Beverage image prompts — drinks and packaging
- Product visual direction AI — define brand style first
- AI-Written Product Copy: Detail Pages, Hero Image Lines, Short-Video Scripts