How to Tell If a Product Listing Photo Is AI-Generated

You found a product on Amazon, Temu, or a secondhand marketplace that looks almost too good. The studio lighting is immaculate, every surface is flawless, and the background is the kind of clean white you’ve only ever seen in luxury brand campaigns. Something feels off — or maybe nothing feels off, and that’s precisely what should worry you.

AI-generated product images are now common enough that major marketplaces have begun issuing policy guidance on them. Sellers use them to avoid photography costs, to show products they haven’t manufactured yet, or — in the deceptive cases — to make goods look better than they are. The same problem extends to real estate listings, where AI-staged rooms and inflated architectural renders routinely misrepresent properties. Knowing the signals is now a basic skill for anyone who buys online.

This guide covers what to look for, why the obvious checks aren’t enough, and how to verify a suspicion for free.

1. Inspect reflections and shadows first

Reflections and shadows are where physics bites AI generators hardest. A real product photograph is taken in a real space with real light sources. A generated render is assembled from learned patterns that approximate physics — and approximation breaks down at the detail level.

Look for:

• Glossy surfaces — does the reflection in a phone screen, lacquered tabletop, or ceramic product match the environment shown? A render might show a reflection of a studio backdrop that isn’t in the image.
• Shadow consistency — all shadows in a real scene fall in the direction of the same light source. If a product casts a shadow to the left and a nearby prop casts one to the right, the scene was assembled rather than photographed.
• Missing shadows — objects resting on surfaces should cast one. A floating-object effect, or a shadow that’s perfectly diffuse when the rest of the lighting suggests a directional source, is a flag.
• Window or lamp reflections in eyewear and screens — if a product includes a screen, lens, or any highly reflective surface, check what’s reflected. AI often fills this with a vague gradient or an implausible environment.

2. Check material textures for impossible uniformity

Real fabrics have pills, weave irregularities, and slight colour variation across the surface. Real leather has grain that changes direction. Real wood has knots and growth-ring patterns that never tile. Real metal has micro-scratches from manufacturing and handling.

AI-generated materials tend to be either too perfect or suspiciously repetitive:

• Fabric where every thread is identical, as if the texture was stamped from one sample
• Leather with a grain that tiles visibly when you zoom in
• Wood grain that ends abruptly or repeats in a way no tree produces
• Metal surfaces without a single machining mark or handling scratch

This is different from professional product photography, which can be retouched. Retouching smooths imperfections but preserves the underlying material structure. Generation creates the surface from scratch — and learned textures have a characteristic regularity.

3. Read the text on labels, packaging and tags

Text is one of the most reliable failure points for image generators. If a product has a label, packaging, care tag, or any printed surface, zoom in as far as the image resolution allows.

Genuine product photos show real text: legible brand names, ingredient lists, regulatory marks, and serial numbers. AI-generated images frequently produce:

• Letter-like shapes that aren’t any real character from any alphabet
• Plausible-looking but meaningless strings (“PROBUCT”, “MAFE IN CHINA” with a transposed letter)
• Text that fades into texture before you can read it
• Logos that resemble a real brand at a glance but fall apart under inspection

This also applies to any incidental text in the scene — books on a staging shelf, signage visible through a window, labels on other props. If you can’t read it, ask why.

4. Examine background edges and staging props

A product sitting in a lifestyle scene — on a desk, on a kitchen counter, styled with complementary objects — should interact with that environment. Look at where the product meets its background:

• Edge blending — an unnatural halo, an overly clean cut, or a slight double-edge where the product outline meets the scene indicates compositing or generation artefacts.
• Props that don’t make physical sense — items that overlap in ways they couldn’t if they were real objects (a book “through” a glass, a fabric drape with no support structure), or props that are identically repeated in different listing photos.
• Backgrounds that are too generic — stock-photo-level environments that could belong to any product in any category, with no location-specific detail.

For real estate specifically: AI-staged rooms often have furniture that hovers slightly above the floor, doesn’t cast proper shadows, or is scaled wrong for the space. Windows showing an idealized view while interior lighting implies a different time of day are a common tell.

5. Cross-reference against other listing photos

Deceptive listings typically mix genuine and AI-generated images. The real photos — often blurry, poorly lit, showing a slightly different product — are buried after the hero shots. Look for:

• Inconsistency in the product itself across photos (different proportions, colour shifts, hardware details that appear or disappear)
• A mix of obviously professional and obviously amateur photography that doesn’t match what you’d expect from the same seller
• Lifestyle images with no real-world reference scale — no hand, no desk, no person — that would let you judge the actual size

On secondhand platforms (eBay, Craigslist, Sahibinden, Letgo, Dolap), the flag is the opposite: stock-perfect images on a platform where real sellers photograph their actual items with their actual phone. If the photo looks like a brand asset, it may be one — or a generated facsimile of one.

Why visual inspection isn’t enough

Everything above is real and worth doing. It’s also beatable. Generators improve constantly, and a well-prompted render with a quality model can pass a careful visual check. Social-media re-compression flattens out the texture irregularities you’d need to see.

What visual inspection can’t do is read the pixel-level data the image is made of. Forensic analysis can:

A genuine product photograph taken with a camera carries the camera’s noise profile in its pixels, EXIF data that names the device, and a natural frequency distribution from the lens and sensor. A generated image has none of these — and forensic tools measure exactly those absences.

How FakeRadar helps

FakeRadar runs a multi-engine forensic check on any uploaded image: an AI-model score built on learned generation signatures, ELA to check compression consistency across regions, FFT spectrum analysis to detect generation grid patterns, and a full metadata pass for EXIF and C2PA provenance records.

The result is reported as signals — not a verdict. FakeRadar tells you whether AI-generation signals were detected and at what strength across each layer. That matters in the product-photo context because some legitimate uses of generated imagery are disclosed, while others misrepresent what the buyer will receive. The evidence lets you make that judgment, rather than a black-box label doing it for you.

For a deeper explanation of what the ELA heatmap shows on a manipulated or generated image, see how to read ELA heatmaps. For why no single score settles the question, see why AI detection is signal-based.

FAQ

Can AI-generated product photos pass a casual glance?

Yes, easily. Modern generators produce renders cleaner than real photography. The tells are physical impossibilities: reflections with no source, contradictory shadows, and textures that are suspiciously uniform.

Do real product photos always have EXIF camera data?

Not always — platforms strip metadata on upload. But AI-generated images often have no camera data to begin with, and forensic analysis detects generation patterns regardless.

Is it legal for sellers to use AI-generated product photos?

Legality varies, but it’s deceptive if the render misrepresents the actual product. Consumer protection laws in most countries cover material misrepresentation regardless of how the image was made.

What’s the most reliable way to check a product listing photo?

Stack multiple signals: visual inspection of reflections, shadows, and textures; reverse image search for stock photo reuse; and forensic AI-detection analysis. No single check is definitive, but several pointing the same direction is meaningful.

Summary

• Reflections, shadows, and material textures are where AI product renders fail at the physical level — look for internal contradictions, not just sloppiness.
• Text on labels, packaging, and props is one of the most reliable visual failure points for generators.
• On secondhand platforms, stock-quality photos are a red flag rather than a reassurance.
• Visual inspection is necessary but not sufficient — forensic analysis reads what your eyes can’t.
• Check any product photo for free when the visual pass leaves you uncertain.