🧠03漫剧系列开发@炮老师的小课堂

Vertical short drama isn't one story - it's an episode machine, and the pacing problem is real: hooking viewers in 60 seconds per episode across 60+ episodes. DapaoDramaDevelop (advanced node 03) is the stage that thinks about that machine before a single line of dialogue gets written. Its whole job is series-level planning: the creative brief, adaptation boundaries, director's direction, the "story engine" (the hook/reward loop that keeps people watching), and an episode map. Its description is deliberate about the boundary: it plans, and it does not rewrite the script.

Where it sits

In the advanced flow it comes right after 原著分析 (Novel analysis, node 02) and before 剧本创作 (script writing, node 04). If you've been through DapaoDramaNovel, the analysis document feeds straight in through the optional 📄 原著分析.md input. You can also skip the novel stage entirely and drive this node from your own brief - it accepts an upstream bundle and the "📝 本阶段要求" box where you describe what you actually want to make.

Think of it as the answer to the question the later nodes are too busy to ask: what's the arc, and which episode is this? When the write node sits down to draft episode EP007, it should already know what has to happen in this episode and what the episode map demands it set up for EP012. That's what Develop locks in, before any prose exists to go stale.

How it works

Like every LLM stage, it runs one (or more) calls against the author's LLM relay at api.dapaoai.com/v1/chat/completions - fill 🔑 API密钥 or leave it blank to use the DAPAO_API_KEY environment variable. The model builds 系列开发.md under a strict authoring contract, and the node runs mechanical checks on the result before it stamps the bundle: the doc has to be structurally sound or downstream refuses it. The output contract is the same as the other analysis stages - a 📦 漫剧资料包, the 📄 阶段文档, and a ℹ️ 处理信息 report (the 🧩 提示词列表 output is empty here; this stage writes no copy-paste prompts).

The inputs that matter

Beyond the shared API-key/model/requirements/seed set, the two things you'll actually touch:

• 📄 原著分析.md - connect your novel analysis here so the planning is grounded in source evidence rather than the model's memory of the genre.
• 📝 本阶段要求 - where you say "this is a sweet revenge story, plan 60 episodes, protagonist needs a public humiliation in ep 1 and full reversal by ep 40." The system prompt leans on the upstream docs and your requirements, and flags anything undecided as 未决 (pending) instead of inventing a creator's approval.

Installing and gotchas

It installs as part of the pack: Manager → "dapaoAPI", or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI into ComfyUI/custom_nodes, then restart and Ctrl+F5. No weights to download - but plan for LLM spend, and remember the fixed-seed 48-hour reuse trick: identical requests don't re-bill, changed ones do.

Two traps worth knowing. First, don't skip this stage and then wonder why your script has no series logic - write (node 04) treats Develop as an upstream dependency and will refuse to build from nothing. Second, if you're here just to plan one episode from an original idea, the guided DapaoDramaPrepare node runs the same planning internally and is a lot less wiring. The advanced branch is for people who want each stage separate, revisable, and individually controllable - which is exactly when this node earns its keep.