WorkBuddy E-commerce Image Task Splitting: Why Does Recognition Get Chaotic?

What annoys me most is having nowhere to put all the assets before making images. I saw that roundup of e-commerce image tools, which said Flux Art can aggregate models. My first reaction was: is there a cheaper alternative? I've used Flux Art for a month; it generates images fine, but my product data, competitor screenshots, white-background shots, and size requirements are still scattered across Shopify, WeChat, and Excel. So I wanted to try WorkBuddy to handle the sorting first, then feed prompts to the image tools.

I've used WorkBuddy for about a month; it was okay for organizing tables before. This time I created a folder with over a dozen images: competitor main images, supplier white-background shots, and two vertical detail page images. I asked it to categorize by "main image, white-background image, detail page," and output platform sizes and prompt drafts. Result: it treated the vertical detail pages as main images, left the size column full of blanks, and added question marks to filenames. I thought it hadn't read all the images, so I tried again with just three images. It identified "white-background image" as "cutout" and told me to run remove.bg.

Let's not calculate cost yet; I'm running out of time. I suspect it reads filenames and notes, barely looking at the actual images. But these images were dragged out of chats, so the names are all gibberish. Does anyone know what WorkBuddy actually reads first when handling e-commerce image tasks? Do I need to manually create a tag column first, or should I let it only organize and not generate image instructions directly? Or is it really worse than me slowly dragging things around in Excel for scenarios with many images?