Open-source 12 e-commerce image-generation Agent Skills

On October 4, 2026, a developer posted on V2EX about an open-source project that packages common e-commerce imaging workflows into 12 Agent Skills. The project is released under the MIT license at https://github.com/xianyu110/ecommerce-image-skills, and the author says the Skills can be dropped straight into Claude Code, Codex, or Cursor.

The 12 Skills cover white-background hero shots, lifestyle scenes, on-model shots, selling-point infographics, A+ modules, size charts, multi-angle sets, promo banners, batch background swaps and SKU color changes, Xiaohongshu covers, Taobao 750 detail-page long images, and a full orchestration skill that plans an entire image set in one pass.

What sets the project apart from a plain prompt collection, the author writes, is that every prompt first locks the product's shape, color, logo, and text, then carries those constraints through generation so the output does not drift away from the actual product. Platform rules are built in as well: Amazon hero images, for example, must use #FFFFFF pure white and keep the subject at 85% or more of the frame, and the repository ships check-and-auto-fix scripts for them.

For generation, the default is GPT Image 2.5 Flare, with a switch to Sunburst when text needs to be preserved or a local edit is required. The post also mentions three ways to generate images, but the original text cuts off while listing those options, so the full set is not given.