6 min read
How AI Checks Product Photo Quality
A clear look at how automated systems judge ecommerce product photos: detecting blur, low resolution, the wrong background, watermarks, multiple products, and bad cropping, and why marketplaces enforce these rules.
Why marketplaces check photos at all
When you upload a product image to a large marketplace, a human almost never looks at it first. An automated pipeline does. It runs in seconds, scores the image against a set of rules, and decides whether the listing goes live, gets suppressed, or gets flagged for review.
The reason is scale. Marketplaces process enormous volumes of new listings, and inconsistent photos hurt the shopping experience. A grid of products looks clean and trustworthy when every main image sits on the same plain background, fills the frame, and is sharp enough to zoom. Poor images lower conversion and erode trust, so platforms turn their style guides into machine-readable checks.
Understanding what these systems look for is practical knowledge. If you know the checks, you can pass them on the first try instead of guessing why a listing was rejected.
Detecting blur and softness
Blur is one of the most common reasons a photo fails. The challenge is that there is usually no perfect reference image to compare against, so systems use what researchers call no-reference image quality assessment: they judge the image on its own.
A widely used technique is the variance of the Laplacian. The Laplacian is a second-derivative filter that responds strongly to edges and fine detail. A sharp, in-focus photo has many crisp edges, so the spread of those responses is wide and the variance is high. A blurry photo has soft, smeared edges, so the variance drops. If the number falls below a tuned threshold, the image is marked as blurry.
Modern pipelines often go further, using machine-learning models trained on many rated images to produce a quality score, sometimes a simple good-or-bad label and sometimes a scale from unusable to excellent. These models can catch motion blur, defocus, noise, and heavy compression artifacts that a single math formula might miss.
Resolution, background, and framing
Resolution is the easiest check because it is just pixel counting. Amazon, for example, asks for at least 1,000 pixels on the longest side and recommends 2,000 or more so the zoom feature works. Anything smaller is flagged before a quality model even runs.
Background checks are stricter than most sellers expect. Amazon main images require a pure white background at RGB 255, 255, 255. Automated systems sample the pixels around the product and measure how far they drift from that exact value. An off-white, cream, or light gray that looks fine to your eye can still be caught, because the algorithm reads the actual numbers.
Framing is checked too. Amazon asks the product to fill about 85 percent or more of the frame. Systems estimate the bounding area of the product against the canvas and flag photos where the item floats in too much empty space or is cropped at the edges. Aspect ratio matters because each marketplace and ad placement expects specific shapes, and a square crop that cuts off part of the product is a common, avoidable failure.
Watermarks, text, and multiple products
Beyond raw quality, systems look for things that should not be in a clean main image. Text and logo detection, often built on optical character recognition and object detection, scans for overlaid words, promotional badges, and watermarks. Marketplaces commonly ban these on the main image because they clutter listings and can signal copied or third-party content.
Object detection also counts what is in the frame. A main image is usually expected to show a single product. If the model detects several distinct items where only one is allowed, or a prop, a mannequin, or a hand where the rules forbid it, the photo can be rejected. The same machinery doubles as safety moderation, catching prohibited or unsafe content.
None of this is magic. Each check is a focused, testable rule: count the pixels, measure the background color, score the sharpness, find the text, count the objects. Stacked together, they approximate the judgment a careful human editor would make, at a speed no human team could match.
What this means for sellers
The takeaway is that good product photos are not only about looking nice; they are about passing a checklist you cannot see. Shoot at high resolution, keep the product sharp and well lit, place it on a genuinely clean background, fill the frame, and strip out any text or watermarks before you upload.
This is where preparing your images ahead of time helps. Renderivo focuses on the visual basics that these checks care about: removing busy backgrounds, producing a clean white background, and framing products in a tidy square. The goal is honest, accurate product photos, not tricks, so your listings clear automated review and look consistent across a marketplace grid.
If you are unsure whether an image will pass, it is worth running it against the specific marketplace rules before you publish, then fixing whatever the checks flag.
Frequently asked questions
Does an AI check every product photo I upload?
On large marketplaces, yes, an automated pipeline reviews images on upload. It measures resolution, background, sharpness, framing, and the presence of text or watermarks, then approves, suppresses, or flags the listing, often before any person sees it.
How does software know if my photo is blurry?
It analyzes the image without a reference. A common method, the variance of the Laplacian, measures the spread of edge detail: sharp photos score high and blurry ones score low. Many systems also use machine-learning models trained on rated images to assign a quality score.
Why was my white background rejected when it looked white to me?
Some marketplaces require a pure white at RGB 255, 255, 255. Automated checks read the exact pixel values, so an off-white, cream, or light gray that looks fine to your eye can still measure as not white and trigger a rejection.
Can I check my images before I list them?
Yes. You can review an image against a marketplace's rules ahead of time and fix issues like low resolution, a non-white background, or stray text. Preparing clean, correctly framed photos first reduces rejections and re-uploads.
Get listing-ready photos the checks will pass
Clean backgrounds, true white, and tidy square framing, so your product photos clear automated review and look consistent. New accounts get free credits.