🎭02漫剧人物与画面@炮老师的小课堂

In the guided three-node 🐠 漫剧 flow (Prepare → Visual → Finish), DapaoDramaVisual is step 02 - the art-direction pass. Feed it the bundle from DapaoDramaPrepare (01) and it does what the advanced chain does in two separate nodes (05 visual assets + 06 image prompts): analyze the script into characters, locations and key props, then write the IMG-* copy-paste image prompts those assets need. The bundle flows through automatically, so there's no branch-and-merge to babysit.

The mode dropdown is the whole node

🎨 处理内容 decides what happens:

• 人物场景分析＋资产图片提示词 - the full pass: analyze assets and write the image prompts.
• 仅人物场景分析 - stop after the visual-setting analysis (视觉设定.md), no prompt writing, if you want to review the breakdown before spending on prompts.
• 仅资产图片提示词（复用已有设定） - skip analysis and just write prompts from a visual setting you already have (e.g. imported or from a previous episode).

Then 👤 资产规划 shapes the asset list: 人物优先 (characters first - the default; the reasoning is that drama lives or dies on the hero's identity board, so it's generated first and everything else is "necessary, not decorative"), 仅人物设定 (characters only), or 完整资产 (characters + scenes + key props). The suite has opinions about not padding: generic doors and walls merge into their scene rather than eating a precious asset slot, and off-screen voices don't get invented faces.

The batch output you'll actually use

🧩 所选资产批量提示词 is the socket that matters - one complete IMG prompt per element, straight to an image node's prompt input, and ComfyUI maps each to its own generation. Two knobs control what's in it:

• 🔢 提示词选择 - leave blank for everything, or filter like 1,3-5 or by full IMG entry ID. Only filters the batch output; never touches the stored document.
• ♻️ 跨集资产输出 (episode 2+) - "全部本集资产" or "仅新增或变化资产". In the only-changes mode, an asset whose prompt is byte-identical to last episode's drops out of the batch (empty list = no new art needed, keep reusing the old reference images). Note "changed" includes mere wording tweaks, so eyeball the doc's 跨集资产变化 section.

🖌️ 图片用途 picks the specialist recipe: identity board, three-view turnaround, look/state variant, location plate, prop plate, composite board, lookdev style comparison, or local edit/revision. Auto means it plans the mix itself.

Money and mechanism

It's an LLM node on the author's relay (api.dapaoai.com/v1/chat/completions): fill 🔑 API密钥 or use DAPAO_API_KEY, pick 🤖 LLM模型. Costs are per internal stage - the full mode is roughly two LLM responses (one analysis, one image-prompts), and switching 处理内容/资产规划 changes the requests, so expect the bill to move. Fixed seed + same inputs = 48-hour reuse; no auto-retry on paid failures. Outputs: 📦 漫剧资料包 (on to Finish), 📄 阶段文档, ℹ️ 处理信息, the batch list, and 💾 完整资料包JSON（续集用）.

Installing and gotchas

Part of the pack: Manager → "dapaoAPI", or clone into custom_nodes; restart + Ctrl+F5. No weights to download.

The recurring honesty note applies here hardest: it writes prompts for images it will never see. The IMG-* prompts are its plan for what you should generate; the docs are explicit that PLAN- entries are not proof any pixel exists, and there's no image input anywhere in this suite. Actual on-screen character consistency is a two-part job - these prompts carry the text-side continuity locks verbatim, but you still generate the reference images and reuse them across episodes yourself. If the "仅新增或变化资产" batch comes back empty on episode 2, that's the suite telling you your existing reference art is still the right art - which is exactly the outcome you want.