🐠Seedance2.0全能视频@炮老师的小课堂

ByteDance's Seedance 2.0 remains one of the strongest video models you'll never get to run locally - the whole line is API-only, no exceptions. DapaoSeedance20AllroundVideoNode ("🐠Seedance2.0全能视频@炮老师的小课堂") is the pack's attempt to give you all of it from one node: text-to-video, image-to-video, first-last-frame interpolation, and full multimodal reference. It's one of the pack's "maintenance" nodes, meaning the author actually keeps this one updated, and it routes through the author's own dapaoAI relay (https://api.dapaoai.com) rather than a third-party reseller.

The modes

Pick from 🎛️ 生成模式:

• 文生视频 - prompt in, video out.
• 图生视频 - an image (or images) drives the shot.
• 首尾帧生视频 - the classic "morph between these two frames" shot. Connect 🎬 首帧图 and 🏁 尾帧图; the first goes in as images[0], the last as the trailing reference.
• 多模态参考 - mix reference images, videos and audio.

The 🤖 模型 dropdown is SD2-face vs SD2 - these are the relay's mapping names. 👤 真人模式 defaults on, which selects SD2-face (the character/mapping model, 720P only). Choose 1080P and the node flips you to SD2 and disables 真人模式 automatically - the tooltips spell this out, and it's a nice touch of guard-railing.

The part that trips people up

Unlike the RunningHub nodes, which accept local media sockets freely, this node's reference materials must be public HTTP/HTTPS URLs. The 🌐 公网素材URL(JSON) field takes a JSON blob shaped {"images":[],"videos":[],"audios":[]} - and the tooltip is explicit: local paths and data URIs are rejected. You'll still find local IMAGE sockets for 首帧图/尾帧图 and the nine 参考图 slots, so it's not URL-everything - but if you're trying to feed a local video or audio file in, the node will not take it. The relay uploads via its own /v1/files endpoint when it can, which is why everything else stays socket-based. The pragmatic pattern for local video: host it, or use the RunningHub nodes instead, which handle local upload natively.

Other settings worth knowing: 🧩 分辨率 (720P/1080P), ⏱️ 时长(秒) 4–15, 📐 视频比例 (16:9 or 9:16 here), 🔊 生成音频, and the poll controls (🔄 最大轮询秒数 default 1800, ⏱️ 轮询间隔, ⌛ 请求超时). 🎲 随机种 is cache-control only - it doesn't go to the API.

Outputs are the pack standard: 🎬 视频 (VIDEO), 🆕 任务ID, 📋 响应信息, 🔗 视频URL.

Install and cost

cd ComfyUI/custom_nodes/
git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git
cd ComfyUI-dapaoAPI
pip install -r requirements.txt

Restart, then register on the dapaoAI platform (sign-up → buy a redeem code in the wallet → apply for an API key) and paste it in. No local models at all.

Now the honest part. Seedance-by-API is pricey - the community's reaction to these nodes being subscription/credit-gated ranged from "extremely expensive" to outright refusal - and this one routes through a relay that sits between you and the model, so you're trusting the author's platform with both your key and your content. It's a "I need this one shot to look great and I don't have 10 minutes of a locally-rendered alternative" node, not a default. For character consistency work, the 真人模式/SD2-face path is the reason to pick this node over a plain API call.