🎉RH全能视频Seedance2.0@炮老师的小课堂

Seedance 2.0 is ByteDance's flagship video model, and it's one of the most requested closed generators of the moment (nearly 900 exact-match corpus mentions) precisely because there are no local weights - the only door is an API. DapaoRHAllVideoSeedanceNode is that door via RunningHub, packaged for ComfyUI: text-to-video, image-to-video, and multimodal video generation through RunningHub's openapi/v2 endpoint, with a VIDEO tensor coming out the other side.

The inputs that matter

• 🌐 API渠道 + 🔑 API密钥 - domestic (runninghub.cn) vs international (runninghub.ai), keys don't cross. Same rule as every RH node in this pack.
• 🤖 模型 - SEEDANCE2.0 (default) and the FAST variant (SEEDANCE2.0-FAST, presumably cheaper/faster, lower quality ceiling).
• 🎛️ 功能 - 文生视频 (T2V), 图生视频 (I2V), 多模态视频.
• 🧩 分辨率 - six options (720p default through higher). ⏱️ 时长(秒) - 13 options from 5s up. 📐 视频比例 - seven, default adaptive.
• 🔊 生成音频 - true by default; Seedance can synth audio. 🧍 真人模式 - true by default. This is the "real person mode" the README advertises (即梦2.0真人模式 lineage) - for character/actor-consistent results.
• 🔄 素材转换槽位 - 15 options, default all. This is a seedance-asset mechanism: which input slots get converted/uploaded for the request.
• 🖼️ 返回尾帧 - whether the job also returns a last-frame image.
• 🔎 联网搜索 - false by default; online search for grounded video generation.
• 🎲 随机种 - default −1. Note: unlike most nodes in this pack, here a −1 seed is genuinely passed through in some RH flows, so it's not strictly cache-only - worth knowing before you hunt for reproducibility.

The optional inputs are the multi-modal heart: 🎬 首帧图, 🏁 尾帧图, 🖼️ 参考图1–9, 🎞️ 参考视频1–3, 🎵 参考音频1–3, plus URL-string inputs (首帧公网URL, 参考图URL列表, etc.) for when your assets are already hosted, an asset_ids string, and the usual 📋 额外参数JSON + polling trio.

Outputs

🎬 视频 (a real VIDEO tensor - save it directly), 🆔 任务ID, 📋 响应信息, 🔗 视频URL, and 🏁 尾帧图 (when 返回尾帧 is on).

How it works

Same async pattern as the RH image nodes: submit, poll (max 1200s, 5s interval), download, decode. The multi-modal bits mean the node has to upload your reference images/video/audio to RunningHub as assets first (that's the 素材转换槽位 plumbing), then the video job references them.

The honest read

Per external-api-nodes.md, the community's split on API video is sharp: the loud center reads "not local, through an API" as a reason to skip, and the cost objection is real - video jobs are where a session gets expensive fast. That's not a reason to avoid Seedance (you can't run it locally, full stop), but it is a reason to watch 时长 and 分辨率 like a hawk. Real-person mode and generated audio are the features people actually pay for here, and RunningHub's price is usually under ByteDance's first-party rate - that's the reseller pitch, and it mostly holds.

Gotchas

• Video bills fast. A 10s 1080p clip costs multiples of a 5s 720p one. Budget before you batch.
• Key mismatch between channels is the classic support question.
• Seed semantics differ from the rest of the pack (default −1, may be passed to the API) - don't assume the cache-only rule applies here.
• Chinese-only labels; README is ad-heavy (invite codes, 送算力) - the author monetizes via referral, par for this ecosystem.

Install: ComfyUI Manager → "dapaoAPI", or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git → pip install -r requirements.txt → restart. RunningHub account + API key required.