Why Your Protein Bar and Your Padel Class Need Completely Different AI

The Black Tile Problem

We kept seeing them during onboarding reviews — mysterious black rectangles where product images should have been. A yoga studio would get a beautiful calendar of posts, but three of the twelve slots would be pitch-black voids. A coffee brand would onboard flawlessly. A tutoring service would produce ghost tiles.

The pattern was subtle enough to miss for weeks. When we finally mapped every black image back to its source, the answer was obvious in retrospect: our pipeline was trying to photograph things that don't physically exist.

A Protein Bar Is Not a Padel Class

Every AI-generated marketing image starts with a reference — a photo of the actual product that the model uses to maintain visual fidelity. For a protein bar, that reference is a crisp shot of the packaging. The AI knows what colors, shapes, and textures to preserve. The output looks like the real product sitting in a real scene.

But what is the "reference photo" for a group fitness class? Or a language tutoring session? Or a monthly subscription box experience? These are services — their value lives in the doing, not in an object you can place on a table. When our pipeline tried to find a product image for these businesses and came up empty, it either generated from nothing (producing black tiles) or hallucinated a physical product that didn't exist.

We needed the AI to understand a fundamental distinction: some products are things you hold, and some are things you do.

Teaching the AI to See the Difference

We introduced a simple but powerful classification at the earliest stage of brand analysis. When our AI first examines a business, it now categorizes every product as either physical or service. The heuristic is intuitive — if you can photograph it sitting on a table and someone would recognize it, it's physical. If the value is the experience itself, it's a service.

This single bit of information cascades through the entire content pipeline. For physical products, the system cranks up visual fidelity to near-maximum, ensuring the generated image faithfully reproduces the actual packaging, colors, and form factor. A snack brand's bar wrapper needs to look exactly right — creative liberties with product appearance destroy trust.

For service businesses, the pipeline pivots entirely. Instead of trying to reproduce a non-existent object, it generates lifestyle and activity imagery. A padel studio gets action shots of players mid-swing. A cooking class gets warm kitchen scenes with engaged participants. The AI focuses on capturing the energy and environment of the experience rather than hunting for a product photo that was never going to exist.

Smarter Calendar Planning

The classification also transformed how we plan content calendars. Physical product businesses now get more hero product shots — about half their posts showcase the actual item, because that's what sells tangible goods. Service businesses get the inverse: sixty percent lifestyle and UGC-style content, with minimal product-only posts, because nobody buys a yoga membership from a photo of a yoga mat.

We also added a safety net: if a product has zero associated images after the full scraping and analysis pipeline, its calendar slots are filtered out entirely rather than generating into the void. No more black tiles, period.

The Scraper Gets Smarter Too

Part of the black image problem traced back even further — to how we collected images in the first place. Instagram-only businesses with no website had zero scraped images to work with. We now fall back to Instagram media as a product image source when web scraping comes up empty. We also expanded our web scraper to try a secondary rendering approach when a site yields fewer than three images, catching JavaScript-heavy single-page applications that our initial pass would miss.

Image downloads themselves got more resilient with automatic retry logic, so transient network failures during onboarding no longer leave gaps in the product image library.

Transparency When Things Are Missing

Even with all these improvements, sometimes a product simply has no usable images — the website is text-only, or the Instagram feed doesn't show the product clearly. Previously, these products would silently disappear from the calendar, leaving the business owner wondering where their main offering went.

Now the pipeline tells you exactly what happened. When a product is dropped because no images could be found, the onboarding response includes a clear list of what was skipped and why. No more mystery gaps in your content calendar.

Faithful Product Compositing

For physical products in influencer-style posts, we also closed a gap in the image generation fallback chain. When our highest-quality generation path fails, the system now creates a face-consistent influencer image and then composites the actual product cutout onto it — preserving both the influencer's likeness and the product's real appearance. Previously, this fallback would generate the influencer but forget the product entirely.

What Changed in the Numbers

The most visible impact is the complete elimination of black tiles in generated calendars. But the subtler win is quality — physical product images now score noticeably higher in visual accuracy because the pipeline knows to prioritize fidelity over creativity. And service businesses get content that actually represents what they sell, rather than awkward AI-imagined objects.

Looking Ahead

This physical-versus-service distinction is just the beginning of product-aware intelligence. We're exploring how the same classification can inform caption tone, hashtag strategy, and even influencer matching — because the person who sells a protein bar and the person who sells a padel membership speak to their audiences in fundamentally different ways.