🎉RH seedance2.0-Mini@炮老师的小课堂

Seedance 2.0 is one of the best video models you cannot run locally - ByteDance keeps the whole Seedance line as an API product, and you either call it or you don't have it. DapaoRHSeedance20MiniNode ("🎉RH seedance2.0-Mini@炮老师的小课堂") is that call, wrapped for ComfyUI and routed through RunningHub. You type a prompt, the node phones RunningHub's SparkVideo 2.0 endpoint, polls the job until it's done, and hands you back a real VIDEO you can wire straight into a preview or save node. From the canvas it looks like any local video generator. It is not. That's the whole point.

How it works

Two modes, selected by 🎛️ 功能:

• 文生视频 (text-to-video): prompt goes in, video comes out.
• 多模态视频 (multimodal): you can additionally feed up to 9 reference images, 3 reference videos, 3 reference audios, and it does image-to-video / reference-to-video style work.

The node is async under the hood - submit a job, poll every few seconds until it's done. That's what the 🔄 最大轮询秒数, ⏱️ 轮询间隔 and ⌛ 请求超时 controls at the bottom are for. Video takes a while; budget the poll window accordingly.

The settings that matter:

• 🖼️ 分辨率 - 480p up to 4K. Higher is slower and costlier.
• ⏱️ 时长(秒) - 4 to 15, or -1 (model decides).
• 👨 真人模式 - this one's interesting. When on, RunningHub tries to convert your materials to asset:// references for stronger character consistency. If you're doing character work, leave it on.
• 🔄 素材转换槽位 - which input slots get converted to assets. Default all.
• 🎲 随机种 - note the difference from local nodes: here it defaults to -1 and it's just for ComfyUI cache control, not the model seed.

The asset system (why it exists)

For face consistency, materials can be uploaded to a RunningHub asset library and referenced by ID - that's what the asset_ids input is for, and it's fed by the pack's other RH nodes (素材/创建 → 素材ID/合并 → here). It's a genuinely useful pipeline for keeping a character consistent across generations: upload once, reference many times. You can also paste raw URLs into the image/video/audio URL list inputs instead.

Outputs

🎬 视频 (VIDEO - connect to a save/preview node), 🆕 任务ID, 📋 响应信息, 🔗 视频URL, and 🏁 尾帧图 (the last frame as an IMAGE, handy for chaining into the next shot).

Installing and the honest cost conversation

Install is pack-standard: ComfyUI Manager → search dapaoAPI, or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git into custom_nodes/, then pip install -r requirements.txt and restart. You need a RunningHub account and API key - the node shows per-second pricing (Seedance 2.0 runs about ¥0.6/second), so it tells you the price right on the node instead of hiding it in a docs page.

That price is the part people underestimate. The community's reaction to Seedance-2.0-as-an-API was immediate and blunt - "extremely expensive," "a really small number of videos per year" for a subscription, and more than one flat "go kick rocks" when it turned out not to be local. The counters are fair too: this is the real price of rendering, and it's the only way to touch a model ByteDance will never open. My take: keep this node for the jobs that genuinely need Seedance - character consistency, cinematic shots, anything local open models embarrass you on - and don't make it your default video path. Also remember your reference images go to a third-party server, so keep the pipeline to material you're comfortable with leaving the machine.