Why does WorkBuddy's screenshot-to-table conversion get messier as it looks more like a table?

Let me say something counter-intuitive: the biggest pitfall in converting images to tables is its eagerness to stitch the image into a "standard table."

I saw Feishu Base's AI paste-to-input feature claiming complex layouts can be restored with one click. My first reaction was: is there an alternative? I've used WorkBuddy for about a month. Organizing product info and after-sales records went smoothly, so I thought I'd use it for image input. Yesterday, I tried three images: a Shopify competitor screenshot, a supplier quote, and a WeChat group order note screenshot. The goal was simple: split into Product Name, Price, Inventory, and Risk Notes.

The result: the quote had merged cells, so it split one product name row into three rows, and prices shifted accordingly. Lineless schedule-style tables were weirder; it said it couldn't recognize the table structure and asked for parsable data. I followed the prompt and reran it, adding a separate instruction to "keep only core fields." Headers appeared, but the order number in the WeChat screenshot was missing a digit. I cross-checked with Excel and Power Query and realized the columns weren't aligned. Let me calculate the savings: manually entering three images takes half an hour, but for batch listing, one wrong row causes customer service and inventory issues simultaneously.

Has anyone encountered this? For WorkBuddy's image-to-table conversion, do you need to crop the image first, or have it only recognize text without generating the table directly? Does throwing too many images at once cause chaos? If you have free alternatives, recommendations please.