Fixing the Bug That Put Random Bedding Ads in My SmartStore Product Videos
When Duvet Ads Invaded My Body Pillow Videos
I was building an automation program to generate short marketing videos for a body pillow. When I checked the rendered output, a summer bedding set banner kept popping up as the background behind the price captions.
I couldn't believe my eyes. When scraping product photos directly from Naver SmartStore (Naver's popular Korean e-commerce marketplace) *sangse peeiji* (detailed product pages), the script had pulled in cross-promotional banners embedded by the seller. The entire collage video and thumbnail set looked like a mess.
Why Image Size Filters Completely Failed
My first thought was to filter out these unwanted images by checking their width, height, or aspect ratio. But the seller's promotional banners used the exact same pixel dimensions as standard product photos.
The script could not tell the difference between an actual body pillow photo and a seasonal duvet advertisement. Even after tweaking my code several times, unrelated promo graphics kept sneaking into the final cut. Out of frustration, I asked Claude how to solve this visual filtering problem.
Catching Promotional Banners by Color Saturation
Claude suggested analyzing the distribution of colors across the images. Authentic body pillow product photos rely mostly on calm, muted tones like beige, wood, and soft neutrals.
Promotional banners, on the other hand, feature bright primary colors for discount badges, sale percentages, and free gift callouts. I calculated the ratio of high-saturation pixels across each image. Standard lifestyle product shots had less than 1% high-saturation pixels, whereas promotional banners easily exceeded 2%.
I also updated the scraper logic to prioritize real buyer review photos first. Real customer photos add authenticity and rarely contain graphic overlays.
Building a Cleaner Automation Pipeline
Adding a color-distribution filter completely resolved the issue. When I ran the scraper and video generator again, all the stray bedding graphics vanished from the thumbnails and videos.
Leading with authentic customer review photos also made the final clips look far more trustworthy. Highly saturated products might occasionally trigger a false positive, but e-commerce listings usually have plenty of buyer photos to fill the gap.
If you scrape Korean shopping pages to create automated media, do not rely solely on image dimensions. Watch out for saturated promotional badges, and let color analysis do the filtering for you.
Common questions
Why do promo banners get scraped from Naver SmartStore pages?
Korean e-commerce sellers frequently insert cross-promotional banners for other products directly into their detailed product descriptions. Web scrapers that gather every embedded image file end up downloading these unrelated promotional graphics alongside actual item photos.
How does color saturation filtering remove unwanted e-commerce banners?
Promotional banners typically use high-saturation red, blue, or yellow badges to highlight discounts and gifts, while real product photos feature neutral, natural lighting. By measuring the percentage of high-saturation pixels, scripts can easily flag and discard promotional graphics.
Why prioritize buyer review photos when building automated video clips?
Buyer review photos show authentic, real-world use without graphic marketing overlays. Using them first avoids promotional banners while making automated marketing content feel more genuine and trustworthy to shoppers.