WorkBuddy Helped Build a Product Copy Alternative; Hold Off on Paid APIs

I came across Baidu Qianfan's smart product marketing copy generation. It claims to identify selling points from product images and combine them with basic info to generate Xiaohongshu (RED) notes, WeChat Moments promotions, and review-style copy. My first reaction was: is this a paid component? Is there a cheaper alternative? Let me calculate the savings for you. Its list price is charged per call and per image, with a daily free quota—it sounds cheap. But for someone like me running a one-person shop doing the work of five, actually integrating it requires setting up an image host, enumerating categories, and cleaning fields. You might get stuck on API field errors before generating even a few pieces of copy.

The Qianfan interface requires category, style, product_info, and img. The img parameter expects an array of downloadable URLs. It’s not just about throwing in an image; you need to organize your product data into a structure machines can read. I’ve been using WorkBuddy for a month. I previously wrote about how it organizes Shopify and Xianyu (Idle Fish) assets. The conclusion is that it handles plain text and tables well, but don’t trust its image content classification too much—it tends to guess based on filenames. So this time, I didn’t let WorkBuddy directly look at images to identify selling points. Instead, I had it first organize local assets into a copywriting workbook ready for listing.

My workflow here is: open WorkBuddy, go to the left-side project panel, create a new project called "Product Copy Pipeline," click Import Local Folder, select 00_Raw_Assets, and write fixed instructions in the task box. Each line represents one SKU, and fields must include category, title, price, color, size, material, selling point keywords, image filename, and image URL. Mark uncertain fields as Pending Manual Confirmation. Don’t let it freestyle at this step. The stricter the fields, the less hassle later.

Then have it generate three sets of candidate copies based on product_info: Xiaohongshu notes, WeChat Moments promos, and product reviews. Each set should be 80–120 characters, ending with three keywords. I limit categories to those supported by Qianfan, such as apparel, food, and home goods, to avoid it writing about cakes as if they were clothes. After generation, export to CSV, then filter using Excel or Power Query.

As a solo operator, my biggest fear is making mistakes today and losing track of versions tomorrow. In my WorkBuddy project, I fix three directories: 00_Raw_Assets is read-only, containing Shopify screenshots, Xianyu order notes, and customer questions from WeChat support; 10_Drafts is editable, holding WorkBuddy-generated copy candidates and tables; 20_Ready_To_List is locked before publishing, containing only confirmed final CSVs and external image links. Raw assets never change, drafts can be messy, but 'Ready To List' must be clean. I run a full batch once a week and only process new SKUs incrementally daily. After WorkBuddy outputs, I manually spot-check 10% of the fields, focusing on color, price, and size. I previously fell into the trap where it identified Red_MainImage as the color. Later, I standardized filenames to SKU_Color_Size_Angle, which stabilized its text field extraction.

If your shop is stable, needs to batch-generate dozens or hundreds of items daily, and has standardized image hosting and SKU tables, components like Qianfan have their value. They save the step from images/data to copy. But for a one-person shop like mine, getting WorkBuddy to smooth out asset organization, field extraction, candidate copy generation, and table exports first is more reliable. Paid APIs are like takeout: fast, but only if your kitchen is already tidy.

Right now, I’m putting Qianfan’s field checklist and WorkBuddy’s table template side-by-side, filling in gaps as needed, without rushing to pay. The time saved goes back to answering customers, packing orders, and taking more photos. Once WorkBuddy truly stabilizes image content recognition, or if Qianfan’s free quota covers my needs for a while, I’ll consider integrating it.