🛍️电商详情页提示词@炮老师的小课堂

If you sell anything on Taobao, Pinduoduo, Douyin, or any Chinese e-commerce platform, you know the grind: a detail page is 8 to 24 long vertical screens, each a separate image with its own copy, call-to-action, and visual continuity, and doing that by hand is a day of work. DapaoDetailFlowPromptNode turns it into one LLM call that emits a per-screen prompt for every screen, so you can pipe them straight into an image generator and get a coherent long-form product page instead of a pile of disconnected pictures.

How it thinks

You describe the product in 📝 原始电商需求 and the node's required inputs do the rest of the merchandising thinking:

• 🗂️ 页面类型 - physical-goods detail page, AI-tool/model product page, SaaS feature page, developer product page, or custom long page.
• 🗂️ 产品品类 - 15 categories (food/drink, beauty, 3C digital, home appliances, apparel, baby, pet, auto, sports, jewelry, AI software…), default auto-detect.
• 🛒 适配平台 - 淘宝/天猫, 京东, 抖音小店, 拼多多, 小红书, 独立站, or 通用长图.
• 🔢 分屏数量 - 1 to 24. This is the dial that defines the page. The tooltip makes a nice promise: whatever the count, it always opens on a first-screen hook and closes on a final-screen CTA, so you never get a page that just… stops.
• 📐 视频比例 wait, no - 页面连续方式: strictly continuous detail page (recommended) vs. unified-visual-but-independent-screens.
• ✍️ 画面文字方案 - how screen text is handled: auto-plan, terse Chinese titles, full copy, only your provided text, reserved blank text zones, or none. And 🔤 必须逐字保留的画面文字 for the bits that can't drift.

The long tail of optional inputs is genuinely useful if you have the data: product name, real specs, target buyer, up to three selling-point directions, evidence type (specs, materials, certifications, user reviews), use scenarios, CTA type, visual style, fact-handling policy (infer-and-mark vs. strict-facts-only vs. marketing-leeway), and up to 12 product reference images plus 6 style reference images. The node's fact-handling is a nice bit of discipline - it flags what's inferred so you don't put fake specs on a real listing.

What comes out

The outputs are the reason to use this over a generic LLM: 🛍️ 第01屏提示词 through 第24屏提示词 - one STRING port per screen, so you can wire screen 3 straight into your generator while screen 7 is still being written. Plus 📋 详情页完整蓝图JSON, 📝 全部准确画面文案, ⚠️ 事实与连续性提醒, 🧩 所选分屏批量提示词 (a list output), 🖼️ 可选视觉母版提示词 (1:3 or 1:4 long-format master prompt), and the usual debug trio.

Two operational warnings from the tooltips, both worth taking literally:

• Long-page JSON is big. The 📝 最大输出令牌 default is 32768, and the tooltip says 9–24 screens should stay there - dropping it can truncate the response, and the node will not auto-retry a paid request. Truncation here means a silently half-built page.
• 📐 提示词适配目标 only rewrites prompt style for the downstream model (GPT Image 2, Banana 2/Pro, FLUX/SD…) - it doesn't change your generator's resolution. You still set 720p-wide-or-whatever in the image node yourself.

The trap

This node produces prompts, not images. It's the perfect front-end for a long-image-capable model, but a generic 1:1 image node will happily mangle a 24-screen page. Pair it with the detail-flow Skill in the chat workbench or with a generator that handles tall aspect ratios, or you'll be re-prompting in a loop.

Install is the pack standard: ComfyUI Manager → "dapaoAPI", or git clone + pip install -r requirements.txt into ComfyUI/custom_nodes/, restart, and paste a dapaoAI key (register at api.dapaoai.com, top up, apply key in the default group). Chinese UI throughout - the whole pack is China-first, which is exactly the right audience for an e-commerce detail-page generator.