You generated a product shot for your Shopify store, your Etsy listing, or a brand campaign — a ceramic mug, a sneaker, a skincare bottle, a watch. The composition looks fine until you look closer: reflections sit wrong on the surface, the shadow does not match the lighting direction, the background is too clean to be a real shoot, the product proportions feel slightly off. Customers do not buy from images that read as “AI.” Fixing it is mostly about lighting setup, material cues, and authentic environmental grounding.
Fastest fix: the single biggest tell is a missing contact shadow plus too-symmetric lighting. Add one concrete lighting setup (key + fill + rim) and one line for a soft contact shadow directly beneath the product, then regenerate. If the material still looks wrong, run image-to-image from a real or phone reference shot instead of text-to-image. The rest of this page is the full diagnosis-to-fix path.
Which tell are you hitting?
Match what you see to the cause and jump to the fix.
| What you see | Most likely cause | Go to |
|---|---|---|
| Product looks like a floating sticker | No contact shadow | Step 2 |
| Surface looks evenly lit, “rendered” | Lighting too symmetric | Step 1 |
| Reflections don’t match any light source | Wrong specular for the material | Step 3 |
| Looks like the wrong material (plastic instead of ceramic) | Material undescribed | Step 3, Step 5 |
| Watch/sneaker proportions feel off | Geometry drift | Step 5 (image-to-image) |
| Logo warped or text garbled | Model can’t render in-image text | Step 7, cause 6 |
| Right product, wrong vibe for the category | Style/category mismatch | Step 4 |
Common causes
Ordered by what makes products read as AI most often.
1. Studio lighting too symmetric
AI defaults to “evenly lit from all sides” because the training data over-represents catalog shots. A real product shoot uses a key light + fill light + rim light combination that creates directional shadows and highlights. AI symmetry reads as “rendered.”
How to spot it: Look at the shadow under and behind the product. If it is evenly faint on all sides, lighting is too symmetric.
2. Reflections wrong for the material
A glossy ceramic mug should show a soft reflection of the light source on its curved surface. AI often paints generic specular highlights that do not match a plausible light setup. Metal products show even more obvious reflection errors (a polished watch case that reflects nothing).
How to spot it: Trace where reflections appear on the product. Do they correspond to where a real light source would be? Are reflections of the studio (other lights, backdrop edges, etc.) consistent across the surface?
3. Background too clean — no shadow, no shoot context
Pure white background with no contact shadow under the product is the dead giveaway. Real catalog shots have at least a soft drop shadow. Lifestyle shots have a real environment.
How to spot it: Is there any shadow connecting the product to the surface it sits on? If no, the product looks like a floating sticker.
4. Material doesn’t behave like the real thing
AI struggles with semi-transparent (frosted glass, white silicone, milky liquids), highly reflective (chrome, mirror polish), and textured-but-soft (suede, fleece, velvet) materials. Outputs often look like a different material than intended.
5. Product proportions / geometry off
Subtle wrong proportions — a watch case slightly elongated, a sneaker heel slightly higher than reality — are detectable to people who know the product category, even if they cannot articulate what is wrong.
6. Brand logo wrong or missing
If your prompt includes a real brand (which most platforms refuse), or asks the model to generate a brand-like logo, the logo comes back warped or with garbled text. Most image models still mangle small in-image text. Two options as of June 2026: hand-edit logos in post (most reliable), or use a model built for legible in-image text — Ideogram v3 or Nano Banana Pro (Gemini 3 Pro Image) render short taglines and clean type far better than general models. For your own real logo, compositing it in Photoshop still beats generating it.
7. Wrong style for the product category
A luxury watch shot with bright cheerful lifestyle styling, or a children’s toy shot with dramatic noir lighting. The category mismatch reads as AI to a customer in that vertical.
Before you change anything
- Find 3-5 real reference photos of products in your category that you consider best-in-class.
- Note the lighting setup, background, and angle for each reference.
- Save your current prompt, model, and the fake-looking output.
- Decide your output use case (e-commerce hero, lifestyle marketing, social) — each tolerates different levels of AI tell.
- Commit or back up the current prompt template before changing it.
Information to collect
- Full prompt, negative prompt, model, version.
- A 100x100 crop of the product’s specular highlights and shadow region.
- 1-3 real reference photos for comparison.
- The category convention (luxury, mass-market, artisanal, sport).
- The intended output channel (Amazon listing, Instagram, brand site hero).
Shortest path to fix
Step 1: Specify a concrete lighting setup
Replace generic descriptors with a setup a real photographer would write:
three-point studio lighting: large softbox key light from upper-left,
fill bounce card on the right, rim light from behind to separate from background,
subtle drop shadow underneathOr for natural light:
natural window light from the left, slight golden warmth,
soft shadow falling to the lower-rightThis single change shifts most “fake studio” outputs to plausible.
Step 2: Add a contact shadow
soft contact shadow directly beneath the product, anchoring it to the surfaceThis single line eliminates the “floating sticker” effect.
Step 3: Describe the material correctly
Generic:
ceramic mugSpecific:
matte unglazed ceramic mug with slight rough texture, hand-thrown imperfections,
soft subsurface light scattering near the rimGeneric:
luxury watchSpecific:
brushed stainless steel watch case with fine grain machining lines,
high-gloss sapphire crystal, soft reflection of overhead softbox visible on the domeThe more specific the material description, the closer the output gets.
Step 4: Match background to category convention
- E-commerce hero: light grey gradient backdrop, soft shadow, 5500K
- Luxury: deep saturated solid color, single key light, dramatic shadow
- Lifestyle: real-context environment (wood table, fabric, soft natural light)
- Tech / clean: pure white seamless with soft shadow
Match the convention of best-in-class brands in your specific category.
Step 5: Use image-to-image with a real product photo
This is the single highest-leverage fix for proportions and material. If you have a real reference photo of a similar product, run image-to-image: on local Stable Diffusion / ComfyUI start around denoise 0.4-0.5 (lower keeps more of the original geometry; raise it only if you need more change). For your own product, photograph it on a phone and use that as the reference.
As of June 2026 the cleanest path is an instruction-based editing model rather than raw denoise tuning. Give a real product photo plus a text instruction to FLUX.1 Kontext [pro], Nano Banana Pro (Gemini 3 Pro Image), or Midjourney’s image reference / editor, and ask it to change only the background or lighting while keeping the product. This preserves real geometry, real reflections, and the real logo — exactly the things text-to-image gets wrong — so you fix the “fake” tells without re-rolling the whole shot.
Step 6: Switch to a product-strong model
Models shift fast; verify the current default before a big batch. Strongest for product imagery as of June 2026:
- FLUX.1.1 Pro for clean catalog shots, glass refraction, and material accuracy (text-to-image realism leader)
- Imagen 4 (Standard for catalog, Ultra for luxury / hero) — top-ranked on specular accuracy and material fidelity for isolated product shots
- Nano Banana Pro (Gemini 3 Pro Image) for editing control, blending a real product into a new scene, and any in-image text or tagline
- Midjourney V8.1 (the default since June 10, 2026) with
--style raw --stylize 100for editorial / luxury feel
Avoid stylized base models (Niji, anime SDXL checkpoints) for product work. No single model wins everything: pros pick per shot — FLUX/Imagen for the hero, an editing model for variants.
Step 7: Post-process to add legitimacy
In Photoshop / Affinity:
- Add a subtle vignette (5% intensity)
- Add film grain (3-5%)
- Sharpen the product slightly while keeping background soft
- Hand-correct logos and text
Even a 60-second pass meaningfully reduces AI tells.
How to confirm the fix
- Zoom 100% on specular highlights — they should match the lighting direction you specified.
- The shadow should fall in a direction consistent with the key light.
- Show the image to someone who knows the product category. Ask whether anything reads as AI.
- Place the image next to your real best-in-class reference. The gap should be narrow, not glaring.
- Cropped to the product alone (no background), does it still hold up?
If it still fails
- Reduce the prompt to the minimum (product + material + one lighting cue), regenerate. Add back specificity one at a time.
- Use a real product photo as the image-to-image starting point — even a phone shot beats text-to-image for product fidelity.
- Switch to a fundamentally different model. Some models genuinely cannot produce e-commerce-grade product imagery yet.
- For luxury or hero placements, photograph the real product instead of generating. AI-generated product imagery still loses against a competent phone shot under window light for many premium use cases.
- Package the prompt, model, output, and references before asking community help.
Prevention
- Use real reference photos whenever possible — phone shots of the actual product are the strongest anchor.
- Save working prompt templates per material type (matte ceramic, brushed metal, glossy plastic, leather, etc.).
- Standardize background and lighting setup per channel (e-commerce, social, lifestyle) so all your assets feel consistent.
- For multi-product catalogs, generate them in one session with the same lighting setup for visual coherence.
- Always run a final post-process pass — even minimal — to add the polish that pure AI output lacks.
FAQ
What is the single biggest reason an AI product image looks fake?
A missing contact shadow. With no shadow connecting the product to the surface, it reads as a floating sticker no matter how good the rest is. Add soft contact shadow directly beneath the product and that one tell usually disappears. The second most common is lighting that is evenly bright on all sides; a directional key + fill + rim setup fixes it.
Is it legal to use AI-generated product images on Amazon, Shopify, or Etsy? The image itself is generally fine, but it must accurately represent the actual product you ship. Amazon and other marketplaces require main listing images to show the real item, so AI hero shots that change shape, color, or features risk a policy violation or returns. Use AI for backgrounds and lifestyle context, and keep at least one true photo of the real product.
Which AI model is best for product photos in 2026?
As of June 2026, FLUX.1.1 Pro and Imagen 4 lead for photorealistic isolated product shots (Imagen 4 ranks highest on specular/material accuracy; Imagen 4 Ultra for luxury). For editing a real product into a new scene or adding legible text, use Nano Banana Pro (Gemini 3 Pro Image) or FLUX.1 Kontext. Midjourney V8.1 with --style raw is strongest for editorial styling. There is no single winner — pick per shot.
Why does my product look like the wrong material (plastic instead of ceramic, etc.)? The prompt is under-specifying the material. “ceramic mug” gives the model latitude to default to generic plastic-like rendering. Describe the surface physics: matte vs glossy, texture, how light scatters or reflects (see Step 3). If it still drifts, run image-to-image from a real photo so the model copies real material behavior instead of inventing it.
Should I just photograph the product myself? For luxury and hero placements, often yes. A competent phone shot under window light with a real contact shadow still beats text-to-image for many premium use cases. The strongest modern workflow is hybrid: shoot the real product once, then use an editing model (FLUX.1 Kontext, Nano Banana Pro) to generate background and lighting variants from that single true photo.
Why are the reflections wrong even though the lighting prompt looks right?
Image models paint plausible-looking specular highlights without simulating a real light setup, so reflections often correspond to no actual source. Either specify exactly where the highlight should appear (e.g. soft reflection of overhead softbox visible on the dome) or, for reflective products like chrome and polished metal, switch to image-to-image from a real photo — reflections are the hardest thing to fake from text alone.