😶‍🌫️Seedance2全能导演@炮老师的小课堂

The hardest part of a video model isn't usually the generation - it's getting the prompt right, and then keeping a multi-clip project coherent shot after shot. DapaoSeedance20DirectorNode ("😶‍🌫️Seedance2全能导演@炮老师的小课堂") is the pack's answer: an LLM "director" that sits upstream of your Seedance video node. You describe the rough idea, it decides what kind of Seedance task you actually need, compiles a finished prompt, and tracks project state across clips so the second shot remembers the first one. It doesn't render anything itself - it's a planning brain that makes the render better.

How it works

It calls an LLM through the dapaoAI relay's /v1/chat/completions endpoint (🤖 LLM模型 - eleven options, default gemini-3.7-flash) and asks it to do a specific job, selected by 🎛️ Seedance任务:

• 自动识别 - the LLM figures out which mode you need from your input.
• Task modes: T2V (text), I2V (image), V2V (video reference), R2V (all-round reference), FLF2V (first-last frame), Edit, Extend (续写), Sequence (连续剧情), Review (复盘), Repair (失败修复).

So you can literally run a clip, feed its failure back, and have the director rewrite the prompt for a retry - that's the "导演" loop, and it's the genuinely clever part of this node.

Inputs that matter

• 📝 原始视频需求 - the only thing you must write well: character, action, camera, sound. The LLM does the rest.
• 🎨 创作类型 - 30 genre presets, from 通用导演 to 极简产品广告, 动作大片, 悬疑惊悚·真相反转, 萌宠治愈日常. This is where you steer the tone.
• 📐 视频比例, 🧩 目标分辨率, ⏱️ 目标时长, 🔊 原生音频 - the target specs baked into the prompt.
• 🌐 输出中文提示词 - off gives English prompts (usually better for the model), on gives Simplified Chinese.
• 📦 上一个项目状态JSON - connect the previous run's 项目状态JSON output here for continuity. This is the feature that makes multi-shot narratives possible: the director knows what already happened.

Optional media: 🎬 首帧图/🏁 尾帧图, up to 9 参考图, 3 参考视频, 3 参考音频 - it can analyze the actual material rather than working blind.

Outputs

Seven strings. The ones you care about: 🎬 Seedance最终提示词 (wire this into your Seedance node's prompt - that's the whole point), 🎛️ 识别任务 (which mode it decided), 📦 项目状态JSON (feed back into the next run), 🧾 当前片段合同 (a per-shot "contract" the LLM commits to), and 📄 LLM完整响应 for debugging.

Install and honest verdict

Same pack install - Manager search dapaoAPI, or clone + pip install -r requirements.txt + restart - and a dapaoAI API key (register, buy a redeem code, mint a key). No models to download.

Is it worth it? If you're making one-off clips, the director is overkill - write your prompt by hand. Where it shines is serialized content: a multi-scene ad, a 5-part mini-story, a character that has to look identical across six clips. That continuity feature (上一轮状态 → 这一轮合同) is something you'd otherwise hand-roll as a giant text-passing workflow. The caveats are the pack's standard ones: it's an LLM API call on top of a video API call, so you're paying twice per clip, and a relay sits between you and the model. Use it for the projects that need the structure, skip it for everything else.