dsh-mxpage

Producing a coherent ecommerce detail page is a long chain: a single product photo must pass through selling-point analysis, hero and section planning, a visual-prompt agent, frame-by-frame generation, inline editing, and batch export. Break any link and the result is a pile of unrelated images. dsh-mxpage ports the full MxPage product-image workbench into the DeepSeek Harness Web profile and forces the model to walk the entire analyze, plan, VPA, and generate pipeline, so the output is a visually consistent detail-page set rather than disconnected one-shots.

The workbench ships eight panels: product-asset management with structured selling-point editing, output configuration and whole-page planning, per-section preview with repaint, enhance, translate, and version control, one-click ZIP or JSON export, a four-step Xiaohongshu carousel flow, batch SKU projects (up to twenty, each with isolated failure handling), an API usage ledger with task retry, and multi-channel diagnostics with connection testing. Twenty mxpage tools cover the full lifecycle from project creation and asset upload through analysis, planning, single-section and whole-page generation, editing, translation, export, and usage queries. Every version is written to disk and never overwritten. Image generation calls a paid OpenAI-compatible Images API; the plugin supports rotating across multiple channels, redacts secrets from logs and session events, and prompts the model to confirm before a whole-page job to avoid surprise quota spend.

If you regularly produce hero and detail images for Taobao, Tmall, JD, or Shopee, build product carousels for Xiaohongshu, or handle dozens of SKU product photos per batch, dsh-mxpage collapses what was once a scatter of prompts, manual edits, and separate export steps into a single DSH conversation window. It is built for ecommerce operators, visual designers, and product teams that deliver product imagery at scale.