📖02漫剧原著分析@炮老师的小课堂

Most of the 🐠 漫剧 suite assumes you're adapting a web novel into a vertical short drama, and DapaoDramaNovel (advanced node 02) is where that novel actually gets read. The first thing to internalize: it needs the text. Type a book title into "📄 原文或现有稿" and the node raises an error - the description is blunt that you can't analyze a novel by name alone. If that sounds pedantic, it's the good kind of pedantic: this whole pack is designed to never quietly fabricate, and that starts here.

What it's for

You reach for it when you have source material and want a grounded adaptation plan rather than vibes: what's worth adapting, the plot's functional beats, character candidates, conflict and continuity threads, and where each claim came from in the original text. Output is the 原著分析.md document (plus a bundle), which the next stage - 系列开发 (series development) - depends on. This is the only stage in the advanced flow with special handling for long source text, because web novels are enormous and an LLM can't swallow one whole.

How it works: three analysis modes

The "🔍 分析方式" dropdown gives you:

• 当前材料分析 - analyze what's in the box once. Cheapest, for short material.
• 抽样快评 - sampled quick review. Reads a spread of segments and reports coverage, explicitly telling you what it didn't read. It won't pretend a sample is a full read.
• 全文分段分析 - the serious mode for long novels. It splits the text by chapter headings using its own indexer (or by plain character count if you pick "🗂️ 原著分段依据 = 按字符分段"), then reads segments up to your "🧮 最大分析段数" cap.

The economics matter. Each segment is one LLM request, plus a summary request on top. If the full text has more segments than your cap, it stops before paying for anything - no surprise bill. If extracted summaries overflow context, it merges them level by level (more calls, reported honestly). Every extracted piece keeps its source character range and the doc appends a traceable # 原文分段提取 appendix, so downstream stages cite real locations instead of re-reading the whole novel into context. Coverage percentage is recorded on the page - nice touch of honesty.

The inputs and outputs that matter

Shared with every LLM stage: 🔑 API密钥 (or leave blank to read DAPAO_API_KEY), 🤖 LLM模型 (defaults to the relay's gemini-3.7-flash), 🌡️ 温度, 📝 最大输出令牌, 🎲 随机种 (default fixed - see below), ⌛ 请求超时. Then the novel-specific set above: 分析方式, 分段依据, 每段字符数 (default 12000), 最大分析段数 (default 16).

Outputs: 📦 漫剧资料包 (bundle - wire into the next stage), 📄 阶段文档 (the analysis text), ℹ️ 处理信息 (usage + a "structure check passed" note), and a 🧩 提示词列表 that will be empty - this stage produces no copy-paste prompts, which confuses people scanning the sockets. That list only fills in for the image/storyboard/video stages.

Installing and running

Same pack, one install: Manager → search "dapaoAPI", or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI into custom_nodes, then restart + Ctrl+F5. No models. The real cost is the LLM: calls go to the author's relay at api.dapaoai.com/v1/chat/completions, billed by usage.

Gotchas

The seed tooltip spells out the clever bit: with a fixed seed, an identical request can be reused from cache for 48 hours - rerunning the same content doesn't necessarily re-bill, but changing anything (or the seed) becomes a new paid request. And there's no auto-retry on paid requests; if one fails, the error surfaces with your key masked. Keep the whole novel out of the shared project config - the node is explicit that source text lives here, not in the bundle sent to every later stage. And if you only need a normal one-shot script, you probably don't need this node at all - that's what the guided DapaoDramaPrepare node is for.