🖼️06漫剧资产图片提示词@炮老师的小课堂

Work your way through the 🐠 漫剧 pipeline and you'll hit a small celebration at DapaoDramaImage (advanced node 06): this is the first node whose 🧩 资产图片提示词列表 output is actually full. Every earlier LLM stage returns an empty prompt list - they write analysis and docs. This one writes the IMG-* entries that become the identity boards, three-views, location plates and prop sheets you'll feed to an actual image generator.

What you're getting

Each IMG-* entry in the 图片提示词.md doc is one full, self-contained copy-paste prompt under a ### 可复制提示词 heading - and "full" means full. A prompt can span multiple >-quoted paragraphs (including model-dialect sections), and the extractor treats the whole quoted block as one prompt. Grab the first line and you'll wonder why the character's outfit changed; the fix is to copy the entire block. That's the single most common beginner mistake with this node.

The node knows about the kinds of production images a drama actually needs and loads specialist recipes per use: character identity boards, three-view turnarounds, look/state variants, location plates, key-prop plates, multi-object composite plates, lookdev style comparisons, and local-edit/revision prompts. Continuity locks from the visual setting (视觉设定.md) must survive verbatim into every applicable prompt - the node mechanically checks, and duplicate prompts are flagged rather than silently dropped.

Inputs that matter

• 📄 视觉设定.md (required) - the assets doc from node 05. Paste it or wire the upstream bundle. No visual setting, no image prompts; the node refuses.
• 📝 本阶段要求 - steer the batch: "hero's face close-up first, then costume turnaround, keep the red scarf identity lock".
• 📚 专项参考 - pick a specialist reference (identity board recipe, three-view production-sheet recipe, etc.) beyond the auto-curated default. More references = more context = watch the budget.
• Shared API key/model/seed/token inputs, as with every LLM stage.

Outputs

📦 漫剧资料包 (bundle with the new doc), 📄 图片提示词文档, the 🧩 资产图片提示词列表 (each element = one complete IMG prompt), and ℹ️ 处理信息. Wire that prompt list to an image node's prompt input and ComfyUI maps each entry to its own generation - the list is deliberately "one prompt per element," no numbering, no wrappers.

Installing and gotchas

Same pack install: Manager → "dapaoAPI", or clone into custom_nodes and restart. No models download. Spend is the dapaoAI LLM relay; fixed-seed + unchanged content = 48-hour reuse, changed content re-bills.

The honest caveat, straight from the docs: IMG- is a plan, not a pixel.* The suite never generates these images and never receives one back - there's no image input anywhere in the pack. So its prompts are written blind: you must actually run them in an image tool, then hang the results where the suite told you to (its PLAN- references in the storyboard stage). Character identity ultimately rides on you reusing those same reference images down the line - the prompt text is only half of the consistency fight. If that workflow sounds like exactly the part you wanted automated, you're not alone; it just isn't this pack's job.